Mercurial > libavcodec.hg
annotate mpegaudioenc.c @ 12475:9fef0a8ddd63 libavcodec
Move mm_support() from libavcodec to libavutil, make it a public
function and rename it to av_get_cpu_flags().
| author | stefano |
|---|---|
| date | Wed, 08 Sep 2010 15:07:14 +0000 |
| parents | dde20597f15e |
| children |
| rev | line source |
|---|---|
| 0 | 1 /* |
| 2 * The simplest mpeg audio layer 2 encoder | |
|
8629
04423b2f6e0b
cosmetics: Remove pointless period after copyright statement non-sentences.
diego
parents:
8595
diff
changeset
|
3 * Copyright (c) 2000, 2001 Fabrice Bellard |
| 0 | 4 * |
|
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
5 * This file is part of FFmpeg. |
|
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
6 * |
|
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
7 * FFmpeg is free software; you can redistribute it and/or |
| 429 | 8 * modify it under the terms of the GNU Lesser General Public |
| 9 * License as published by the Free Software Foundation; either | |
|
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
10 * version 2.1 of the License, or (at your option) any later version. |
| 0 | 11 * |
|
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
12 * FFmpeg is distributed in the hope that it will be useful, |
| 0 | 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 429 | 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 15 * Lesser General Public License for more details. | |
| 0 | 16 * |
| 429 | 17 * You should have received a copy of the GNU Lesser General Public |
|
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
18 * License along with FFmpeg; if not, write to the Free Software |
|
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
2979
diff
changeset
|
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 0 | 20 */ |
| 2967 | 21 |
| 1106 | 22 /** |
|
11644
7dd2a45249a9
Remove explicit filename from Doxygen @file commands.
diego
parents:
11560
diff
changeset
|
23 * @file |
| 1106 | 24 * The simplest mpeg audio layer 2 encoder. |
| 25 */ | |
| 2967 | 26 |
| 64 | 27 #include "avcodec.h" |
|
9411
4cb7c65fc775
Split bitstream.h, put the bitstream writer stuff in the new file
stefano
parents:
8718
diff
changeset
|
28 #include "put_bits.h" |
| 8595 | 29 |
| 30 #undef CONFIG_MPEGAUDIO_HP | |
| 31 #define CONFIG_MPEGAUDIO_HP 0 | |
| 0 | 32 #include "mpegaudio.h" |
| 33 | |
|
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
34 /* currently, cannot change these constants (need to modify |
|
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
35 quantization stage) */ |
| 1064 | 36 #define MUL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS) |
| 84 | 37 |
| 38 #define SAMPLES_BUF_SIZE 4096 | |
| 39 | |
| 40 typedef struct MpegAudioContext { | |
| 41 PutBitContext pb; | |
| 42 int nb_channels; | |
| 43 int lsf; /* 1 if mpeg2 low bitrate selected */ | |
| 44 int bitrate_index; /* bit rate */ | |
| 45 int freq_index; | |
| 46 int frame_size; /* frame size, in bits, without padding */ | |
| 47 /* padding computation */ | |
| 48 int frame_frac, frame_frac_incr, do_padding; | |
| 49 short samples_buf[MPA_MAX_CHANNELS][SAMPLES_BUF_SIZE]; /* buffer for filter */ | |
| 50 int samples_offset[MPA_MAX_CHANNELS]; /* offset in samples_buf */ | |
| 51 int sb_samples[MPA_MAX_CHANNELS][3][12][SBLIMIT]; | |
| 52 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3]; /* scale factors */ | |
| 53 /* code to group 3 scale factors */ | |
| 2967 | 54 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT]; |
| 84 | 55 int sblimit; /* number of used subbands */ |
| 56 const unsigned char *alloc_table; | |
| 57 } MpegAudioContext; | |
| 58 | |
| 0 | 59 /* define it to use floats in quantization (I don't like floats !) */ |
|
10328
d4c97368f3e4
Use floating point mathematics when encoding mpeg audio.
cehoyos
parents:
10145
diff
changeset
|
60 #define USE_FLOATS |
| 0 | 61 |
|
5031
70f194a2ee53
move some common mpeg audio tables from mpegaudiodectab.h to mpegaudiodata.c
aurel
parents:
4885
diff
changeset
|
62 #include "mpegaudiodata.h" |
| 0 | 63 #include "mpegaudiotab.h" |
| 64 | |
|
6517
48759bfbd073
Apply 'cold' attribute to init/uninit functions in libavcodec
zuxy
parents:
5161
diff
changeset
|
65 static av_cold int MPA_encode_init(AVCodecContext *avctx) |
| 0 | 66 { |
| 67 MpegAudioContext *s = avctx->priv_data; | |
| 68 int freq = avctx->sample_rate; | |
| 69 int bitrate = avctx->bit_rate; | |
| 70 int channels = avctx->channels; | |
| 84 | 71 int i, v, table; |
| 0 | 72 float a; |
| 73 | |
|
4885
4f351b1e02bc
check for channels<=0 and print a reasonable error message
alex
parents:
4472
diff
changeset
|
74 if (channels <= 0 || channels > 2){ |
|
4f351b1e02bc
check for channels<=0 and print a reasonable error message
alex
parents:
4472
diff
changeset
|
75 av_log(avctx, AV_LOG_ERROR, "encoding %d channel(s) is not allowed in mp2\n", channels); |
| 0 | 76 return -1; |
|
4885
4f351b1e02bc
check for channels<=0 and print a reasonable error message
alex
parents:
4472
diff
changeset
|
77 } |
| 0 | 78 bitrate = bitrate / 1000; |
| 79 s->nb_channels = channels; | |
| 80 avctx->frame_size = MPA_FRAME_SIZE; | |
| 81 | |
| 82 /* encoding freq */ | |
| 83 s->lsf = 0; | |
| 84 for(i=0;i<3;i++) { | |
| 5032 | 85 if (ff_mpa_freq_tab[i] == freq) |
| 0 | 86 break; |
| 5032 | 87 if ((ff_mpa_freq_tab[i] / 2) == freq) { |
| 0 | 88 s->lsf = 1; |
| 89 break; | |
| 90 } | |
| 91 } | |
| 2124 | 92 if (i == 3){ |
| 93 av_log(avctx, AV_LOG_ERROR, "Sampling rate %d is not allowed in mp2\n", freq); | |
| 0 | 94 return -1; |
| 2124 | 95 } |
| 0 | 96 s->freq_index = i; |
| 97 | |
| 98 /* encoding bitrate & frequency */ | |
| 99 for(i=0;i<15;i++) { | |
| 5032 | 100 if (ff_mpa_bitrate_tab[s->lsf][1][i] == bitrate) |
| 0 | 101 break; |
| 102 } | |
| 2124 | 103 if (i == 15){ |
| 104 av_log(avctx, AV_LOG_ERROR, "bitrate %d is not allowed in mp2\n", bitrate); | |
| 0 | 105 return -1; |
| 2124 | 106 } |
| 0 | 107 s->bitrate_index = i; |
| 108 | |
| 109 /* compute total header size & pad bit */ | |
| 2967 | 110 |
| 0 | 111 a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0); |
| 112 s->frame_size = ((int)a) * 8; | |
| 113 | |
| 114 /* frame fractional size to compute padding */ | |
| 115 s->frame_frac = 0; | |
| 116 s->frame_frac_incr = (int)((a - floor(a)) * 65536.0); | |
| 2967 | 117 |
| 0 | 118 /* select the right allocation table */ |
|
5052
d981eb275c8f
remove dependency of mpeg audio encoder over mpeg audio decoder
aurel
parents:
5032
diff
changeset
|
119 table = ff_mpa_l2_select_table(bitrate, s->nb_channels, freq, s->lsf); |
| 84 | 120 |
| 0 | 121 /* number of used subbands */ |
| 5032 | 122 s->sblimit = ff_mpa_sblimit_table[table]; |
| 123 s->alloc_table = ff_mpa_alloc_tables[table]; | |
| 0 | 124 |
|
9999
c78fd9154378
Change av_log() calls surrounded by '#ifdef DEBUG' into dprintf macros.
diego
parents:
9431
diff
changeset
|
125 dprintf(avctx, "%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n", |
|
c78fd9154378
Change av_log() calls surrounded by '#ifdef DEBUG' into dprintf macros.
diego
parents:
9431
diff
changeset
|
126 bitrate, freq, s->frame_size, table, s->frame_frac_incr); |
| 0 | 127 |
| 128 for(i=0;i<s->nb_channels;i++) | |
| 129 s->samples_offset[i] = 0; | |
| 130 | |
| 84 | 131 for(i=0;i<257;i++) { |
| 132 int v; | |
| 5032 | 133 v = ff_mpa_enwindow[i]; |
|
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
134 #if WFRAC_BITS != 16 |
|
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
135 v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS); |
|
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
136 #endif |
| 84 | 137 filter_bank[i] = v; |
| 138 if ((i & 63) != 0) | |
| 139 v = -v; | |
| 140 if (i != 0) | |
| 141 filter_bank[512 - i] = v; | |
| 0 | 142 } |
| 84 | 143 |
| 0 | 144 for(i=0;i<64;i++) { |
| 145 v = (int)(pow(2.0, (3 - i) / 3.0) * (1 << 20)); | |
| 146 if (v <= 0) | |
| 147 v = 1; | |
| 148 scale_factor_table[i] = v; | |
| 149 #ifdef USE_FLOATS | |
| 150 scale_factor_inv_table[i] = pow(2.0, -(3 - i) / 3.0) / (float)(1 << 20); | |
| 151 #else | |
| 152 #define P 15 | |
| 153 scale_factor_shift[i] = 21 - P - (i / 3); | |
| 154 scale_factor_mult[i] = (1 << P) * pow(2.0, (i % 3) / 3.0); | |
| 155 #endif | |
| 156 } | |
| 157 for(i=0;i<128;i++) { | |
| 158 v = i - 64; | |
| 159 if (v <= -3) | |
| 160 v = 0; | |
| 161 else if (v < 0) | |
| 162 v = 1; | |
| 163 else if (v == 0) | |
| 164 v = 2; | |
| 165 else if (v < 3) | |
| 166 v = 3; | |
| 2967 | 167 else |
| 0 | 168 v = 4; |
| 169 scale_diff_table[i] = v; | |
| 170 } | |
| 171 | |
| 172 for(i=0;i<17;i++) { | |
| 5032 | 173 v = ff_mpa_quant_bits[i]; |
| 2967 | 174 if (v < 0) |
| 0 | 175 v = -v; |
| 176 else | |
| 177 v = v * 3; | |
| 178 total_quant_bits[i] = 12 * v; | |
| 179 } | |
| 180 | |
| 925 | 181 avctx->coded_frame= avcodec_alloc_frame(); |
| 182 avctx->coded_frame->key_frame= 1; | |
| 183 | |
| 0 | 184 return 0; |
| 185 } | |
| 186 | |
| 84 | 187 /* 32 point floating point IDCT without 1/sqrt(2) coef zero scaling */ |
|
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
188 static void idct32(int *out, int *tab) |
| 0 | 189 { |
| 190 int i, j; | |
| 191 int *t, *t1, xr; | |
| 192 const int *xp = costab32; | |
| 193 | |
| 194 for(j=31;j>=3;j-=2) tab[j] += tab[j - 2]; | |
| 2967 | 195 |
| 0 | 196 t = tab + 30; |
| 197 t1 = tab + 2; | |
| 198 do { | |
| 199 t[0] += t[-4]; | |
| 200 t[1] += t[1 - 4]; | |
| 201 t -= 4; | |
| 202 } while (t != t1); | |
| 203 | |
| 204 t = tab + 28; | |
| 205 t1 = tab + 4; | |
| 206 do { | |
| 207 t[0] += t[-8]; | |
| 208 t[1] += t[1-8]; | |
| 209 t[2] += t[2-8]; | |
| 210 t[3] += t[3-8]; | |
| 211 t -= 8; | |
| 212 } while (t != t1); | |
| 2967 | 213 |
| 0 | 214 t = tab; |
| 215 t1 = tab + 32; | |
| 216 do { | |
| 2967 | 217 t[ 3] = -t[ 3]; |
| 218 t[ 6] = -t[ 6]; | |
| 219 | |
| 220 t[11] = -t[11]; | |
| 221 t[12] = -t[12]; | |
| 222 t[13] = -t[13]; | |
| 223 t[15] = -t[15]; | |
| 0 | 224 t += 16; |
| 225 } while (t != t1); | |
| 226 | |
| 2967 | 227 |
| 0 | 228 t = tab; |
| 229 t1 = tab + 8; | |
| 230 do { | |
| 231 int x1, x2, x3, x4; | |
| 2967 | 232 |
| 0 | 233 x3 = MUL(t[16], FIX(SQRT2*0.5)); |
| 234 x4 = t[0] - x3; | |
| 235 x3 = t[0] + x3; | |
| 2967 | 236 |
| 0 | 237 x2 = MUL(-(t[24] + t[8]), FIX(SQRT2*0.5)); |
| 238 x1 = MUL((t[8] - x2), xp[0]); | |
| 239 x2 = MUL((t[8] + x2), xp[1]); | |
| 240 | |
| 241 t[ 0] = x3 + x1; | |
| 242 t[ 8] = x4 - x2; | |
| 243 t[16] = x4 + x2; | |
| 244 t[24] = x3 - x1; | |
| 245 t++; | |
| 246 } while (t != t1); | |
| 247 | |
| 248 xp += 2; | |
| 249 t = tab; | |
| 250 t1 = tab + 4; | |
| 251 do { | |
| 252 xr = MUL(t[28],xp[0]); | |
| 253 t[28] = (t[0] - xr); | |
| 254 t[0] = (t[0] + xr); | |
| 255 | |
| 256 xr = MUL(t[4],xp[1]); | |
| 257 t[ 4] = (t[24] - xr); | |
| 258 t[24] = (t[24] + xr); | |
| 2967 | 259 |
| 0 | 260 xr = MUL(t[20],xp[2]); |
| 261 t[20] = (t[8] - xr); | |
| 262 t[ 8] = (t[8] + xr); | |
| 2967 | 263 |
| 0 | 264 xr = MUL(t[12],xp[3]); |
| 265 t[12] = (t[16] - xr); | |
| 266 t[16] = (t[16] + xr); | |
| 267 t++; | |
| 268 } while (t != t1); | |
| 269 xp += 4; | |
| 270 | |
| 271 for (i = 0; i < 4; i++) { | |
| 272 xr = MUL(tab[30-i*4],xp[0]); | |
| 273 tab[30-i*4] = (tab[i*4] - xr); | |
| 274 tab[ i*4] = (tab[i*4] + xr); | |
| 2967 | 275 |
| 0 | 276 xr = MUL(tab[ 2+i*4],xp[1]); |
| 277 tab[ 2+i*4] = (tab[28-i*4] - xr); | |
| 278 tab[28-i*4] = (tab[28-i*4] + xr); | |
| 2967 | 279 |
| 0 | 280 xr = MUL(tab[31-i*4],xp[0]); |
| 281 tab[31-i*4] = (tab[1+i*4] - xr); | |
| 282 tab[ 1+i*4] = (tab[1+i*4] + xr); | |
| 2967 | 283 |
| 0 | 284 xr = MUL(tab[ 3+i*4],xp[1]); |
| 285 tab[ 3+i*4] = (tab[29-i*4] - xr); | |
| 286 tab[29-i*4] = (tab[29-i*4] + xr); | |
| 2967 | 287 |
| 0 | 288 xp += 2; |
| 289 } | |
| 290 | |
| 291 t = tab + 30; | |
| 292 t1 = tab + 1; | |
| 293 do { | |
| 294 xr = MUL(t1[0], *xp); | |
| 295 t1[0] = (t[0] - xr); | |
| 296 t[0] = (t[0] + xr); | |
| 297 t -= 2; | |
| 298 t1 += 2; | |
| 299 xp++; | |
| 300 } while (t >= tab); | |
| 301 | |
| 302 for(i=0;i<32;i++) { | |
|
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
303 out[i] = tab[bitinv32[i]]; |
| 0 | 304 } |
| 305 } | |
| 306 | |
|
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
307 #define WSHIFT (WFRAC_BITS + 15 - FRAC_BITS) |
|
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
308 |
|
12262
dde20597f15e
Use "const" qualifier for pointers that point to input data of
reimar
parents:
12034
diff
changeset
|
309 static void filter(MpegAudioContext *s, int ch, const short *samples, int incr) |
| 0 | 310 { |
| 311 short *p, *q; | |
|
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
312 int sum, offset, i, j; |
|
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
313 int tmp[64]; |
| 0 | 314 int tmp1[32]; |
| 315 int *out; | |
| 316 | |
| 317 // print_pow1(samples, 1152); | |
| 318 | |
| 319 offset = s->samples_offset[ch]; | |
| 320 out = &s->sb_samples[ch][0][0][0]; | |
| 321 for(j=0;j<36;j++) { | |
| 322 /* 32 samples at once */ | |
| 323 for(i=0;i<32;i++) { | |
| 324 s->samples_buf[ch][offset + (31 - i)] = samples[0]; | |
| 325 samples += incr; | |
| 326 } | |
| 327 | |
| 328 /* filter */ | |
| 329 p = s->samples_buf[ch] + offset; | |
| 330 q = filter_bank; | |
| 331 /* maxsum = 23169 */ | |
| 332 for(i=0;i<64;i++) { | |
| 333 sum = p[0*64] * q[0*64]; | |
| 334 sum += p[1*64] * q[1*64]; | |
| 335 sum += p[2*64] * q[2*64]; | |
| 336 sum += p[3*64] * q[3*64]; | |
| 337 sum += p[4*64] * q[4*64]; | |
| 338 sum += p[5*64] * q[5*64]; | |
| 339 sum += p[6*64] * q[6*64]; | |
| 340 sum += p[7*64] * q[7*64]; | |
|
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
341 tmp[i] = sum; |
| 0 | 342 p++; |
| 343 q++; | |
| 344 } | |
|
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
345 tmp1[0] = tmp[16] >> WSHIFT; |
|
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
346 for( i=1; i<=16; i++ ) tmp1[i] = (tmp[i+16]+tmp[16-i]) >> WSHIFT; |
|
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
347 for( i=17; i<=31; i++ ) tmp1[i] = (tmp[i+16]-tmp[80-i]) >> WSHIFT; |
| 0 | 348 |
|
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
349 idct32(out, tmp1); |
| 0 | 350 |
| 351 /* advance of 32 samples */ | |
| 352 offset -= 32; | |
| 353 out += 32; | |
| 354 /* handle the wrap around */ | |
| 355 if (offset < 0) { | |
| 2967 | 356 memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32), |
| 0 | 357 s->samples_buf[ch], (512 - 32) * 2); |
| 358 offset = SAMPLES_BUF_SIZE - 512; | |
| 359 } | |
| 360 } | |
| 361 s->samples_offset[ch] = offset; | |
| 362 | |
| 363 // print_pow(s->sb_samples, 1152); | |
| 364 } | |
| 365 | |
| 366 static void compute_scale_factors(unsigned char scale_code[SBLIMIT], | |
| 2967 | 367 unsigned char scale_factors[SBLIMIT][3], |
| 0 | 368 int sb_samples[3][12][SBLIMIT], |
| 369 int sblimit) | |
| 370 { | |
| 371 int *p, vmax, v, n, i, j, k, code; | |
| 372 int index, d1, d2; | |
| 373 unsigned char *sf = &scale_factors[0][0]; | |
| 2967 | 374 |
| 0 | 375 for(j=0;j<sblimit;j++) { |
| 376 for(i=0;i<3;i++) { | |
| 377 /* find the max absolute value */ | |
| 378 p = &sb_samples[i][0][j]; | |
| 379 vmax = abs(*p); | |
| 380 for(k=1;k<12;k++) { | |
| 381 p += SBLIMIT; | |
| 382 v = abs(*p); | |
| 383 if (v > vmax) | |
| 384 vmax = v; | |
| 385 } | |
| 386 /* compute the scale factor index using log 2 computations */ | |
| 6961 | 387 if (vmax > 1) { |
| 70 | 388 n = av_log2(vmax); |
| 2967 | 389 /* n is the position of the MSB of vmax. now |
| 0 | 390 use at most 2 compares to find the index */ |
| 391 index = (21 - n) * 3 - 3; | |
| 392 if (index >= 0) { | |
| 393 while (vmax <= scale_factor_table[index+1]) | |
| 394 index++; | |
| 395 } else { | |
| 396 index = 0; /* very unlikely case of overflow */ | |
| 397 } | |
| 398 } else { | |
|
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
399 index = 62; /* value 63 is not allowed */ |
| 0 | 400 } |
|
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
401 |
| 0 | 402 #if 0 |
| 2967 | 403 printf("%2d:%d in=%x %x %d\n", |
| 0 | 404 j, i, vmax, scale_factor_table[index], index); |
| 405 #endif | |
| 406 /* store the scale factor */ | |
| 407 assert(index >=0 && index <= 63); | |
| 408 sf[i] = index; | |
| 409 } | |
| 410 | |
| 411 /* compute the transmission factor : look if the scale factors | |
| 412 are close enough to each other */ | |
| 413 d1 = scale_diff_table[sf[0] - sf[1] + 64]; | |
| 414 d2 = scale_diff_table[sf[1] - sf[2] + 64]; | |
| 2967 | 415 |
| 0 | 416 /* handle the 25 cases */ |
| 417 switch(d1 * 5 + d2) { | |
| 418 case 0*5+0: | |
| 419 case 0*5+4: | |
| 420 case 3*5+4: | |
| 421 case 4*5+0: | |
| 422 case 4*5+4: | |
| 423 code = 0; | |
| 424 break; | |
| 425 case 0*5+1: | |
| 426 case 0*5+2: | |
| 427 case 4*5+1: | |
| 428 case 4*5+2: | |
| 429 code = 3; | |
| 430 sf[2] = sf[1]; | |
| 431 break; | |
| 432 case 0*5+3: | |
| 433 case 4*5+3: | |
| 434 code = 3; | |
| 435 sf[1] = sf[2]; | |
| 436 break; | |
| 437 case 1*5+0: | |
| 438 case 1*5+4: | |
| 439 case 2*5+4: | |
| 440 code = 1; | |
| 441 sf[1] = sf[0]; | |
| 442 break; | |
| 443 case 1*5+1: | |
| 444 case 1*5+2: | |
| 445 case 2*5+0: | |
| 446 case 2*5+1: | |
| 447 case 2*5+2: | |
| 448 code = 2; | |
| 449 sf[1] = sf[2] = sf[0]; | |
| 450 break; | |
| 451 case 2*5+3: | |
| 452 case 3*5+3: | |
| 453 code = 2; | |
| 454 sf[0] = sf[1] = sf[2]; | |
| 455 break; | |
| 456 case 3*5+0: | |
| 457 case 3*5+1: | |
| 458 case 3*5+2: | |
| 459 code = 2; | |
| 460 sf[0] = sf[2] = sf[1]; | |
| 461 break; | |
| 462 case 1*5+3: | |
| 463 code = 2; | |
| 464 if (sf[0] > sf[2]) | |
| 465 sf[0] = sf[2]; | |
| 466 sf[1] = sf[2] = sf[0]; | |
| 467 break; | |
| 468 default: | |
| 5127 | 469 assert(0); //cannot happen |
|
2522
e25782262d7d
kill warnings patch by (M?ns Rullg?rd <mru inprovide com>)
michael
parents:
2398
diff
changeset
|
470 code = 0; /* kill warning */ |
| 0 | 471 } |
| 2967 | 472 |
| 0 | 473 #if 0 |
| 2967 | 474 printf("%d: %2d %2d %2d %d %d -> %d\n", j, |
| 0 | 475 sf[0], sf[1], sf[2], d1, d2, code); |
| 476 #endif | |
| 477 scale_code[j] = code; | |
| 478 sf += 3; | |
| 479 } | |
| 480 } | |
| 481 | |
| 482 /* The most important function : psycho acoustic module. In this | |
| 483 encoder there is basically none, so this is the worst you can do, | |
| 484 but also this is the simpler. */ | |
| 485 static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT]) | |
| 486 { | |
| 487 int i; | |
| 488 | |
| 489 for(i=0;i<s->sblimit;i++) { | |
| 490 smr[i] = (int)(fixed_smr[i] * 10); | |
| 491 } | |
| 492 } | |
| 493 | |
| 494 | |
| 495 #define SB_NOTALLOCATED 0 | |
| 496 #define SB_ALLOCATED 1 | |
| 497 #define SB_NOMORE 2 | |
| 498 | |
| 499 /* Try to maximize the smr while using a number of bits inferior to | |
| 500 the frame size. I tried to make the code simpler, faster and | |
| 501 smaller than other encoders :-) */ | |
| 2967 | 502 static void compute_bit_allocation(MpegAudioContext *s, |
| 0 | 503 short smr1[MPA_MAX_CHANNELS][SBLIMIT], |
| 504 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], | |
| 505 int *padding) | |
| 506 { | |
| 507 int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size; | |
| 508 int incr; | |
| 509 short smr[MPA_MAX_CHANNELS][SBLIMIT]; | |
| 510 unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT]; | |
| 511 const unsigned char *alloc; | |
| 512 | |
| 513 memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT); | |
| 514 memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT); | |
| 515 memset(bit_alloc, 0, s->nb_channels * SBLIMIT); | |
| 2967 | 516 |
| 0 | 517 /* compute frame size and padding */ |
| 518 max_frame_size = s->frame_size; | |
| 519 s->frame_frac += s->frame_frac_incr; | |
| 520 if (s->frame_frac >= 65536) { | |
| 521 s->frame_frac -= 65536; | |
| 522 s->do_padding = 1; | |
| 523 max_frame_size += 8; | |
| 524 } else { | |
| 525 s->do_padding = 0; | |
| 526 } | |
| 527 | |
| 528 /* compute the header + bit alloc size */ | |
| 529 current_frame_size = 32; | |
| 530 alloc = s->alloc_table; | |
| 531 for(i=0;i<s->sblimit;i++) { | |
| 532 incr = alloc[0]; | |
| 533 current_frame_size += incr * s->nb_channels; | |
| 534 alloc += 1 << incr; | |
| 535 } | |
| 536 for(;;) { | |
| 537 /* look for the subband with the largest signal to mask ratio */ | |
| 538 max_sb = -1; | |
| 539 max_ch = -1; | |
| 6929 | 540 max_smr = INT_MIN; |
| 0 | 541 for(ch=0;ch<s->nb_channels;ch++) { |
| 542 for(i=0;i<s->sblimit;i++) { | |
| 543 if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) { | |
| 544 max_smr = smr[ch][i]; | |
| 545 max_sb = i; | |
| 546 max_ch = ch; | |
| 547 } | |
| 548 } | |
| 549 } | |
| 550 #if 0 | |
| 2967 | 551 printf("current=%d max=%d max_sb=%d alloc=%d\n", |
| 0 | 552 current_frame_size, max_frame_size, max_sb, |
| 553 bit_alloc[max_sb]); | |
| 2967 | 554 #endif |
| 0 | 555 if (max_sb < 0) |
| 556 break; | |
| 2967 | 557 |
| 0 | 558 /* find alloc table entry (XXX: not optimal, should use |
| 559 pointer table) */ | |
| 560 alloc = s->alloc_table; | |
| 561 for(i=0;i<max_sb;i++) { | |
| 562 alloc += 1 << alloc[0]; | |
| 563 } | |
| 564 | |
| 565 if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) { | |
| 566 /* nothing was coded for this band: add the necessary bits */ | |
| 567 incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6; | |
| 568 incr += total_quant_bits[alloc[1]]; | |
| 569 } else { | |
| 570 /* increments bit allocation */ | |
| 571 b = bit_alloc[max_ch][max_sb]; | |
| 2967 | 572 incr = total_quant_bits[alloc[b + 1]] - |
| 0 | 573 total_quant_bits[alloc[b]]; |
| 574 } | |
| 575 | |
| 576 if (current_frame_size + incr <= max_frame_size) { | |
| 577 /* can increase size */ | |
| 578 b = ++bit_alloc[max_ch][max_sb]; | |
| 579 current_frame_size += incr; | |
| 580 /* decrease smr by the resolution we added */ | |
| 581 smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]]; | |
| 582 /* max allocation size reached ? */ | |
| 583 if (b == ((1 << alloc[0]) - 1)) | |
| 584 subband_status[max_ch][max_sb] = SB_NOMORE; | |
| 585 else | |
| 586 subband_status[max_ch][max_sb] = SB_ALLOCATED; | |
| 587 } else { | |
| 588 /* cannot increase the size of this subband */ | |
| 589 subband_status[max_ch][max_sb] = SB_NOMORE; | |
| 590 } | |
| 591 } | |
| 592 *padding = max_frame_size - current_frame_size; | |
| 593 assert(*padding >= 0); | |
| 594 | |
| 595 #if 0 | |
| 596 for(i=0;i<s->sblimit;i++) { | |
| 597 printf("%d ", bit_alloc[i]); | |
| 598 } | |
| 599 printf("\n"); | |
| 600 #endif | |
| 601 } | |
| 602 | |
| 603 /* | |
| 604 * Output the mpeg audio layer 2 frame. Note how the code is small | |
| 605 * compared to other encoders :-) | |
| 606 */ | |
| 607 static void encode_frame(MpegAudioContext *s, | |
| 608 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], | |
| 609 int padding) | |
| 610 { | |
| 611 int i, j, k, l, bit_alloc_bits, b, ch; | |
| 612 unsigned char *sf; | |
| 613 int q[3]; | |
| 614 PutBitContext *p = &s->pb; | |
| 615 | |
| 616 /* header */ | |
| 617 | |
| 618 put_bits(p, 12, 0xfff); | |
| 619 put_bits(p, 1, 1 - s->lsf); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */ | |
| 620 put_bits(p, 2, 4-2); /* layer 2 */ | |
| 621 put_bits(p, 1, 1); /* no error protection */ | |
| 622 put_bits(p, 4, s->bitrate_index); | |
| 623 put_bits(p, 2, s->freq_index); | |
| 624 put_bits(p, 1, s->do_padding); /* use padding */ | |
| 625 put_bits(p, 1, 0); /* private_bit */ | |
| 626 put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO); | |
| 627 put_bits(p, 2, 0); /* mode_ext */ | |
| 628 put_bits(p, 1, 0); /* no copyright */ | |
| 629 put_bits(p, 1, 1); /* original */ | |
| 630 put_bits(p, 2, 0); /* no emphasis */ | |
| 631 | |
| 632 /* bit allocation */ | |
| 633 j = 0; | |
| 634 for(i=0;i<s->sblimit;i++) { | |
| 635 bit_alloc_bits = s->alloc_table[j]; | |
| 636 for(ch=0;ch<s->nb_channels;ch++) { | |
| 637 put_bits(p, bit_alloc_bits, bit_alloc[ch][i]); | |
| 638 } | |
| 639 j += 1 << bit_alloc_bits; | |
| 640 } | |
| 2967 | 641 |
| 0 | 642 /* scale codes */ |
| 643 for(i=0;i<s->sblimit;i++) { | |
| 644 for(ch=0;ch<s->nb_channels;ch++) { | |
| 2967 | 645 if (bit_alloc[ch][i]) |
| 0 | 646 put_bits(p, 2, s->scale_code[ch][i]); |
| 647 } | |
| 648 } | |
| 649 | |
| 650 /* scale factors */ | |
| 651 for(i=0;i<s->sblimit;i++) { | |
| 652 for(ch=0;ch<s->nb_channels;ch++) { | |
| 653 if (bit_alloc[ch][i]) { | |
| 654 sf = &s->scale_factors[ch][i][0]; | |
| 655 switch(s->scale_code[ch][i]) { | |
| 656 case 0: | |
| 657 put_bits(p, 6, sf[0]); | |
| 658 put_bits(p, 6, sf[1]); | |
| 659 put_bits(p, 6, sf[2]); | |
| 660 break; | |
| 661 case 3: | |
| 662 case 1: | |
| 663 put_bits(p, 6, sf[0]); | |
| 664 put_bits(p, 6, sf[2]); | |
| 665 break; | |
| 666 case 2: | |
| 667 put_bits(p, 6, sf[0]); | |
| 668 break; | |
| 669 } | |
| 670 } | |
| 671 } | |
| 672 } | |
| 2967 | 673 |
| 0 | 674 /* quantization & write sub band samples */ |
| 675 | |
| 676 for(k=0;k<3;k++) { | |
| 677 for(l=0;l<12;l+=3) { | |
| 678 j = 0; | |
| 679 for(i=0;i<s->sblimit;i++) { | |
| 680 bit_alloc_bits = s->alloc_table[j]; | |
| 681 for(ch=0;ch<s->nb_channels;ch++) { | |
| 682 b = bit_alloc[ch][i]; | |
| 683 if (b) { | |
| 684 int qindex, steps, m, sample, bits; | |
| 685 /* we encode 3 sub band samples of the same sub band at a time */ | |
| 686 qindex = s->alloc_table[j+b]; | |
| 5032 | 687 steps = ff_mpa_quant_steps[qindex]; |
| 0 | 688 for(m=0;m<3;m++) { |
| 689 sample = s->sb_samples[ch][k][l + m][i]; | |
| 690 /* divide by scale factor */ | |
| 691 #ifdef USE_FLOATS | |
| 692 { | |
| 693 float a; | |
| 694 a = (float)sample * scale_factor_inv_table[s->scale_factors[ch][i][k]]; | |
| 695 q[m] = (int)((a + 1.0) * steps * 0.5); | |
| 696 } | |
| 697 #else | |
| 698 { | |
| 699 int q1, e, shift, mult; | |
| 700 e = s->scale_factors[ch][i][k]; | |
| 701 shift = scale_factor_shift[e]; | |
| 702 mult = scale_factor_mult[e]; | |
| 2967 | 703 |
| 0 | 704 /* normalize to P bits */ |
| 705 if (shift < 0) | |
| 706 q1 = sample << (-shift); | |
| 707 else | |
| 708 q1 = sample >> shift; | |
| 709 q1 = (q1 * mult) >> P; | |
| 710 q[m] = ((q1 + (1 << P)) * steps) >> (P + 1); | |
| 711 } | |
| 712 #endif | |
| 713 if (q[m] >= steps) | |
| 714 q[m] = steps - 1; | |
| 715 assert(q[m] >= 0 && q[m] < steps); | |
| 716 } | |
| 5032 | 717 bits = ff_mpa_quant_bits[qindex]; |
| 0 | 718 if (bits < 0) { |
| 719 /* group the 3 values to save bits */ | |
| 2967 | 720 put_bits(p, -bits, |
| 0 | 721 q[0] + steps * (q[1] + steps * q[2])); |
| 722 #if 0 | |
| 2967 | 723 printf("%d: gr1 %d\n", |
| 0 | 724 i, q[0] + steps * (q[1] + steps * q[2])); |
| 725 #endif | |
| 726 } else { | |
| 727 #if 0 | |
| 2967 | 728 printf("%d: gr3 %d %d %d\n", |
| 0 | 729 i, q[0], q[1], q[2]); |
| 2967 | 730 #endif |
| 0 | 731 put_bits(p, bits, q[0]); |
| 732 put_bits(p, bits, q[1]); | |
| 733 put_bits(p, bits, q[2]); | |
| 734 } | |
| 735 } | |
| 736 } | |
| 737 /* next subband in alloc table */ | |
| 2967 | 738 j += 1 << bit_alloc_bits; |
| 0 | 739 } |
| 740 } | |
| 741 } | |
| 742 | |
| 743 /* padding */ | |
| 744 for(i=0;i<padding;i++) | |
| 745 put_bits(p, 1, 0); | |
| 746 | |
| 747 /* flush */ | |
| 748 flush_put_bits(p); | |
| 749 } | |
| 750 | |
| 1057 | 751 static int MPA_encode_frame(AVCodecContext *avctx, |
| 2979 | 752 unsigned char *frame, int buf_size, void *data) |
| 0 | 753 { |
| 754 MpegAudioContext *s = avctx->priv_data; | |
|
12262
dde20597f15e
Use "const" qualifier for pointers that point to input data of
reimar
parents:
12034
diff
changeset
|
755 const short *samples = data; |
| 0 | 756 short smr[MPA_MAX_CHANNELS][SBLIMIT]; |
| 757 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT]; | |
| 758 int padding, i; | |
| 759 | |
| 760 for(i=0;i<s->nb_channels;i++) { | |
| 761 filter(s, i, samples + i, s->nb_channels); | |
| 762 } | |
| 763 | |
| 764 for(i=0;i<s->nb_channels;i++) { | |
| 2967 | 765 compute_scale_factors(s->scale_code[i], s->scale_factors[i], |
| 0 | 766 s->sb_samples[i], s->sblimit); |
| 767 } | |
| 768 for(i=0;i<s->nb_channels;i++) { | |
| 769 psycho_acoustic_model(s, smr[i]); | |
| 770 } | |
| 771 compute_bit_allocation(s, smr, bit_alloc, &padding); | |
| 772 | |
|
1522
79dddc5cd990
removed the obsolete and unused parameters of init_put_bits
alex
parents:
1106
diff
changeset
|
773 init_put_bits(&s->pb, frame, MPA_MAX_CODED_FRAME_SIZE); |
| 0 | 774 |
| 775 encode_frame(s, bit_alloc, padding); | |
| 2967 | 776 |
| 9431 | 777 return put_bits_ptr(&s->pb) - s->pb.buf; |
| 0 | 778 } |
| 779 | |
|
6517
48759bfbd073
Apply 'cold' attribute to init/uninit functions in libavcodec
zuxy
parents:
5161
diff
changeset
|
780 static av_cold int MPA_encode_close(AVCodecContext *avctx) |
| 925 | 781 { |
| 782 av_freep(&avctx->coded_frame); | |
|
1031
19de1445beb2
use av_malloc() functions - added av_strdup and av_realloc()
bellard
parents:
925
diff
changeset
|
783 return 0; |
| 925 | 784 } |
| 0 | 785 |
| 786 AVCodec mp2_encoder = { | |
| 787 "mp2", | |
|
11560
8a4984c5cacc
Define AVMediaType enum, and use it instead of enum CodecType, which
stefano
parents:
10328
diff
changeset
|
788 AVMEDIA_TYPE_AUDIO, |
| 0 | 789 CODEC_ID_MP2, |
| 790 sizeof(MpegAudioContext), | |
| 791 MPA_encode_init, | |
| 792 MPA_encode_frame, | |
| 925 | 793 MPA_encode_close, |
| 0 | 794 NULL, |
|
10145
7955db355703
Make sample_fmts and channel_layouts compound literals const to reduce size of
reimar
parents:
9999
diff
changeset
|
795 .sample_fmts = (const enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE}, |
|
11654
fe105c6d28c7
Set .supported_samplerates for mpeg audio encoders.
michael
parents:
11644
diff
changeset
|
796 .supported_samplerates= (const int[]){44100, 48000, 32000, 22050, 24000, 16000, 0}, |
|
7040
e943e1409077
Make AVCodec long_names definition conditional depending on CONFIG_SMALL.
stefano
parents:
6961
diff
changeset
|
797 .long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"), |
| 0 | 798 }; |
|
440
000aeeac27a2
* started to cleanup name clashes for onetime compilation
kabi
parents:
429
diff
changeset
|
799 |
|
000aeeac27a2
* started to cleanup name clashes for onetime compilation
kabi
parents:
429
diff
changeset
|
800 #undef FIX |
