Mercurial > libavcodec.hg
annotate mpegaudio.c @ 3485:f574901e49d2 libavcodec
some fixmes if someone ever wants to implement theora support correctly ...
author | michael |
---|---|
date | Sun, 16 Jul 2006 20:47:48 +0000 |
parents | 0b546eab515d |
children | c8c591fe26f8 |
rev | line source |
---|---|
0 | 1 /* |
2 * The simplest mpeg audio layer 2 encoder | |
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 | |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
2979
diff
changeset
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
0 | 18 */ |
2967 | 19 |
1106 | 20 /** |
21 * @file mpegaudio.c | |
22 * The simplest mpeg audio layer 2 encoder. | |
23 */ | |
2967 | 24 |
64 | 25 #include "avcodec.h" |
2398
582e635cfa08
common.c -> bitstream.c (and the single non bitstream func -> utils.c)
michael
parents:
2281
diff
changeset
|
26 #include "bitstream.h" |
0 | 27 #include "mpegaudio.h" |
28 | |
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
29 /* 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
|
30 quantization stage) */ |
1064 | 31 #define MUL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS) |
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
32 #define FIX(a) ((int)((a) * (1 << FRAC_BITS))) |
84 | 33 |
34 #define SAMPLES_BUF_SIZE 4096 | |
35 | |
36 typedef struct MpegAudioContext { | |
37 PutBitContext pb; | |
38 int nb_channels; | |
39 int freq, bit_rate; | |
40 int lsf; /* 1 if mpeg2 low bitrate selected */ | |
41 int bitrate_index; /* bit rate */ | |
42 int freq_index; | |
43 int frame_size; /* frame size, in bits, without padding */ | |
1064 | 44 int64_t nb_samples; /* total number of samples encoded */ |
84 | 45 /* padding computation */ |
46 int frame_frac, frame_frac_incr, do_padding; | |
47 short samples_buf[MPA_MAX_CHANNELS][SAMPLES_BUF_SIZE]; /* buffer for filter */ | |
48 int samples_offset[MPA_MAX_CHANNELS]; /* offset in samples_buf */ | |
49 int sb_samples[MPA_MAX_CHANNELS][3][12][SBLIMIT]; | |
50 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3]; /* scale factors */ | |
51 /* code to group 3 scale factors */ | |
2967 | 52 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT]; |
84 | 53 int sblimit; /* number of used subbands */ |
54 const unsigned char *alloc_table; | |
55 } MpegAudioContext; | |
56 | |
0 | 57 /* define it to use floats in quantization (I don't like floats !) */ |
58 //#define USE_FLOATS | |
59 | |
60 #include "mpegaudiotab.h" | |
61 | |
1057 | 62 static int MPA_encode_init(AVCodecContext *avctx) |
0 | 63 { |
64 MpegAudioContext *s = avctx->priv_data; | |
65 int freq = avctx->sample_rate; | |
66 int bitrate = avctx->bit_rate; | |
67 int channels = avctx->channels; | |
84 | 68 int i, v, table; |
0 | 69 float a; |
70 | |
71 if (channels > 2) | |
72 return -1; | |
73 bitrate = bitrate / 1000; | |
74 s->nb_channels = channels; | |
75 s->freq = freq; | |
76 s->bit_rate = bitrate * 1000; | |
77 avctx->frame_size = MPA_FRAME_SIZE; | |
78 | |
79 /* encoding freq */ | |
80 s->lsf = 0; | |
81 for(i=0;i<3;i++) { | |
2967 | 82 if (mpa_freq_tab[i] == freq) |
0 | 83 break; |
84 | 84 if ((mpa_freq_tab[i] / 2) == freq) { |
0 | 85 s->lsf = 1; |
86 break; | |
87 } | |
88 } | |
2124 | 89 if (i == 3){ |
90 av_log(avctx, AV_LOG_ERROR, "Sampling rate %d is not allowed in mp2\n", freq); | |
0 | 91 return -1; |
2124 | 92 } |
0 | 93 s->freq_index = i; |
94 | |
95 /* encoding bitrate & frequency */ | |
96 for(i=0;i<15;i++) { | |
2967 | 97 if (mpa_bitrate_tab[s->lsf][1][i] == bitrate) |
0 | 98 break; |
99 } | |
2124 | 100 if (i == 15){ |
101 av_log(avctx, AV_LOG_ERROR, "bitrate %d is not allowed in mp2\n", bitrate); | |
0 | 102 return -1; |
2124 | 103 } |
0 | 104 s->bitrate_index = i; |
105 | |
106 /* compute total header size & pad bit */ | |
2967 | 107 |
0 | 108 a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0); |
109 s->frame_size = ((int)a) * 8; | |
110 | |
111 /* frame fractional size to compute padding */ | |
112 s->frame_frac = 0; | |
113 s->frame_frac_incr = (int)((a - floor(a)) * 65536.0); | |
2967 | 114 |
0 | 115 /* select the right allocation table */ |
84 | 116 table = l2_select_table(bitrate, s->nb_channels, freq, s->lsf); |
117 | |
0 | 118 /* number of used subbands */ |
119 s->sblimit = sblimit_table[table]; | |
120 s->alloc_table = alloc_tables[table]; | |
121 | |
122 #ifdef DEBUG | |
2967 | 123 av_log(avctx, AV_LOG_DEBUG, "%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n", |
0 | 124 bitrate, freq, s->frame_size, table, s->frame_frac_incr); |
125 #endif | |
126 | |
127 for(i=0;i<s->nb_channels;i++) | |
128 s->samples_offset[i] = 0; | |
129 | |
84 | 130 for(i=0;i<257;i++) { |
131 int v; | |
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
132 v = mpa_enwindow[i]; |
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
133 #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
|
134 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
|
135 #endif |
84 | 136 filter_bank[i] = v; |
137 if ((i & 63) != 0) | |
138 v = -v; | |
139 if (i != 0) | |
140 filter_bank[512 - i] = v; | |
0 | 141 } |
84 | 142 |
0 | 143 for(i=0;i<64;i++) { |
144 v = (int)(pow(2.0, (3 - i) / 3.0) * (1 << 20)); | |
145 if (v <= 0) | |
146 v = 1; | |
147 scale_factor_table[i] = v; | |
148 #ifdef USE_FLOATS | |
149 scale_factor_inv_table[i] = pow(2.0, -(3 - i) / 3.0) / (float)(1 << 20); | |
150 #else | |
151 #define P 15 | |
152 scale_factor_shift[i] = 21 - P - (i / 3); | |
153 scale_factor_mult[i] = (1 << P) * pow(2.0, (i % 3) / 3.0); | |
154 #endif | |
155 } | |
156 for(i=0;i<128;i++) { | |
157 v = i - 64; | |
158 if (v <= -3) | |
159 v = 0; | |
160 else if (v < 0) | |
161 v = 1; | |
162 else if (v == 0) | |
163 v = 2; | |
164 else if (v < 3) | |
165 v = 3; | |
2967 | 166 else |
0 | 167 v = 4; |
168 scale_diff_table[i] = v; | |
169 } | |
170 | |
171 for(i=0;i<17;i++) { | |
172 v = quant_bits[i]; | |
2967 | 173 if (v < 0) |
0 | 174 v = -v; |
175 else | |
176 v = v * 3; | |
177 total_quant_bits[i] = 12 * v; | |
178 } | |
179 | |
925 | 180 avctx->coded_frame= avcodec_alloc_frame(); |
181 avctx->coded_frame->key_frame= 1; | |
182 | |
0 | 183 return 0; |
184 } | |
185 | |
84 | 186 /* 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
|
187 static void idct32(int *out, int *tab) |
0 | 188 { |
189 int i, j; | |
190 int *t, *t1, xr; | |
191 const int *xp = costab32; | |
192 | |
193 for(j=31;j>=3;j-=2) tab[j] += tab[j - 2]; | |
2967 | 194 |
0 | 195 t = tab + 30; |
196 t1 = tab + 2; | |
197 do { | |
198 t[0] += t[-4]; | |
199 t[1] += t[1 - 4]; | |
200 t -= 4; | |
201 } while (t != t1); | |
202 | |
203 t = tab + 28; | |
204 t1 = tab + 4; | |
205 do { | |
206 t[0] += t[-8]; | |
207 t[1] += t[1-8]; | |
208 t[2] += t[2-8]; | |
209 t[3] += t[3-8]; | |
210 t -= 8; | |
211 } while (t != t1); | |
2967 | 212 |
0 | 213 t = tab; |
214 t1 = tab + 32; | |
215 do { | |
2967 | 216 t[ 3] = -t[ 3]; |
217 t[ 6] = -t[ 6]; | |
218 | |
219 t[11] = -t[11]; | |
220 t[12] = -t[12]; | |
221 t[13] = -t[13]; | |
222 t[15] = -t[15]; | |
0 | 223 t += 16; |
224 } while (t != t1); | |
225 | |
2967 | 226 |
0 | 227 t = tab; |
228 t1 = tab + 8; | |
229 do { | |
230 int x1, x2, x3, x4; | |
2967 | 231 |
0 | 232 x3 = MUL(t[16], FIX(SQRT2*0.5)); |
233 x4 = t[0] - x3; | |
234 x3 = t[0] + x3; | |
2967 | 235 |
0 | 236 x2 = MUL(-(t[24] + t[8]), FIX(SQRT2*0.5)); |
237 x1 = MUL((t[8] - x2), xp[0]); | |
238 x2 = MUL((t[8] + x2), xp[1]); | |
239 | |
240 t[ 0] = x3 + x1; | |
241 t[ 8] = x4 - x2; | |
242 t[16] = x4 + x2; | |
243 t[24] = x3 - x1; | |
244 t++; | |
245 } while (t != t1); | |
246 | |
247 xp += 2; | |
248 t = tab; | |
249 t1 = tab + 4; | |
250 do { | |
251 xr = MUL(t[28],xp[0]); | |
252 t[28] = (t[0] - xr); | |
253 t[0] = (t[0] + xr); | |
254 | |
255 xr = MUL(t[4],xp[1]); | |
256 t[ 4] = (t[24] - xr); | |
257 t[24] = (t[24] + xr); | |
2967 | 258 |
0 | 259 xr = MUL(t[20],xp[2]); |
260 t[20] = (t[8] - xr); | |
261 t[ 8] = (t[8] + xr); | |
2967 | 262 |
0 | 263 xr = MUL(t[12],xp[3]); |
264 t[12] = (t[16] - xr); | |
265 t[16] = (t[16] + xr); | |
266 t++; | |
267 } while (t != t1); | |
268 xp += 4; | |
269 | |
270 for (i = 0; i < 4; i++) { | |
271 xr = MUL(tab[30-i*4],xp[0]); | |
272 tab[30-i*4] = (tab[i*4] - xr); | |
273 tab[ i*4] = (tab[i*4] + xr); | |
2967 | 274 |
0 | 275 xr = MUL(tab[ 2+i*4],xp[1]); |
276 tab[ 2+i*4] = (tab[28-i*4] - xr); | |
277 tab[28-i*4] = (tab[28-i*4] + xr); | |
2967 | 278 |
0 | 279 xr = MUL(tab[31-i*4],xp[0]); |
280 tab[31-i*4] = (tab[1+i*4] - xr); | |
281 tab[ 1+i*4] = (tab[1+i*4] + xr); | |
2967 | 282 |
0 | 283 xr = MUL(tab[ 3+i*4],xp[1]); |
284 tab[ 3+i*4] = (tab[29-i*4] - xr); | |
285 tab[29-i*4] = (tab[29-i*4] + xr); | |
2967 | 286 |
0 | 287 xp += 2; |
288 } | |
289 | |
290 t = tab + 30; | |
291 t1 = tab + 1; | |
292 do { | |
293 xr = MUL(t1[0], *xp); | |
294 t1[0] = (t[0] - xr); | |
295 t[0] = (t[0] + xr); | |
296 t -= 2; | |
297 t1 += 2; | |
298 xp++; | |
299 } while (t >= tab); | |
300 | |
301 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
|
302 out[i] = tab[bitinv32[i]]; |
0 | 303 } |
304 } | |
305 | |
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
306 #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
|
307 |
0 | 308 static void filter(MpegAudioContext *s, int ch, short *samples, int incr) |
309 { | |
310 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
|
311 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
|
312 int tmp[64]; |
0 | 313 int tmp1[32]; |
314 int *out; | |
315 | |
316 // print_pow1(samples, 1152); | |
317 | |
318 offset = s->samples_offset[ch]; | |
319 out = &s->sb_samples[ch][0][0][0]; | |
320 for(j=0;j<36;j++) { | |
321 /* 32 samples at once */ | |
322 for(i=0;i<32;i++) { | |
323 s->samples_buf[ch][offset + (31 - i)] = samples[0]; | |
324 samples += incr; | |
325 } | |
326 | |
327 /* filter */ | |
328 p = s->samples_buf[ch] + offset; | |
329 q = filter_bank; | |
330 /* maxsum = 23169 */ | |
331 for(i=0;i<64;i++) { | |
332 sum = p[0*64] * q[0*64]; | |
333 sum += p[1*64] * q[1*64]; | |
334 sum += p[2*64] * q[2*64]; | |
335 sum += p[3*64] * q[3*64]; | |
336 sum += p[4*64] * q[4*64]; | |
337 sum += p[5*64] * q[5*64]; | |
338 sum += p[6*64] * q[6*64]; | |
339 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
|
340 tmp[i] = sum; |
0 | 341 p++; |
342 q++; | |
343 } | |
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
344 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
|
345 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
|
346 for( i=17; i<=31; i++ ) tmp1[i] = (tmp[i+16]-tmp[80-i]) >> WSHIFT; |
0 | 347 |
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
348 idct32(out, tmp1); |
0 | 349 |
350 /* advance of 32 samples */ | |
351 offset -= 32; | |
352 out += 32; | |
353 /* handle the wrap around */ | |
354 if (offset < 0) { | |
2967 | 355 memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32), |
0 | 356 s->samples_buf[ch], (512 - 32) * 2); |
357 offset = SAMPLES_BUF_SIZE - 512; | |
358 } | |
359 } | |
360 s->samples_offset[ch] = offset; | |
361 | |
362 // print_pow(s->sb_samples, 1152); | |
363 } | |
364 | |
365 static void compute_scale_factors(unsigned char scale_code[SBLIMIT], | |
2967 | 366 unsigned char scale_factors[SBLIMIT][3], |
0 | 367 int sb_samples[3][12][SBLIMIT], |
368 int sblimit) | |
369 { | |
370 int *p, vmax, v, n, i, j, k, code; | |
371 int index, d1, d2; | |
372 unsigned char *sf = &scale_factors[0][0]; | |
2967 | 373 |
0 | 374 for(j=0;j<sblimit;j++) { |
375 for(i=0;i<3;i++) { | |
376 /* find the max absolute value */ | |
377 p = &sb_samples[i][0][j]; | |
378 vmax = abs(*p); | |
379 for(k=1;k<12;k++) { | |
380 p += SBLIMIT; | |
381 v = abs(*p); | |
382 if (v > vmax) | |
383 vmax = v; | |
384 } | |
385 /* compute the scale factor index using log 2 computations */ | |
386 if (vmax > 0) { | |
70 | 387 n = av_log2(vmax); |
2967 | 388 /* n is the position of the MSB of vmax. now |
0 | 389 use at most 2 compares to find the index */ |
390 index = (21 - n) * 3 - 3; | |
391 if (index >= 0) { | |
392 while (vmax <= scale_factor_table[index+1]) | |
393 index++; | |
394 } else { | |
395 index = 0; /* very unlikely case of overflow */ | |
396 } | |
397 } else { | |
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
398 index = 62; /* value 63 is not allowed */ |
0 | 399 } |
89
2e88e3afecd0
corrected mpeg audio encoding overflows - now it should give correct quality even for very high volumes
glantau
parents:
84
diff
changeset
|
400 |
0 | 401 #if 0 |
2967 | 402 printf("%2d:%d in=%x %x %d\n", |
0 | 403 j, i, vmax, scale_factor_table[index], index); |
404 #endif | |
405 /* store the scale factor */ | |
406 assert(index >=0 && index <= 63); | |
407 sf[i] = index; | |
408 } | |
409 | |
410 /* compute the transmission factor : look if the scale factors | |
411 are close enough to each other */ | |
412 d1 = scale_diff_table[sf[0] - sf[1] + 64]; | |
413 d2 = scale_diff_table[sf[1] - sf[2] + 64]; | |
2967 | 414 |
0 | 415 /* handle the 25 cases */ |
416 switch(d1 * 5 + d2) { | |
417 case 0*5+0: | |
418 case 0*5+4: | |
419 case 3*5+4: | |
420 case 4*5+0: | |
421 case 4*5+4: | |
422 code = 0; | |
423 break; | |
424 case 0*5+1: | |
425 case 0*5+2: | |
426 case 4*5+1: | |
427 case 4*5+2: | |
428 code = 3; | |
429 sf[2] = sf[1]; | |
430 break; | |
431 case 0*5+3: | |
432 case 4*5+3: | |
433 code = 3; | |
434 sf[1] = sf[2]; | |
435 break; | |
436 case 1*5+0: | |
437 case 1*5+4: | |
438 case 2*5+4: | |
439 code = 1; | |
440 sf[1] = sf[0]; | |
441 break; | |
442 case 1*5+1: | |
443 case 1*5+2: | |
444 case 2*5+0: | |
445 case 2*5+1: | |
446 case 2*5+2: | |
447 code = 2; | |
448 sf[1] = sf[2] = sf[0]; | |
449 break; | |
450 case 2*5+3: | |
451 case 3*5+3: | |
452 code = 2; | |
453 sf[0] = sf[1] = sf[2]; | |
454 break; | |
455 case 3*5+0: | |
456 case 3*5+1: | |
457 case 3*5+2: | |
458 code = 2; | |
459 sf[0] = sf[2] = sf[1]; | |
460 break; | |
461 case 1*5+3: | |
462 code = 2; | |
463 if (sf[0] > sf[2]) | |
464 sf[0] = sf[2]; | |
465 sf[1] = sf[2] = sf[0]; | |
466 break; | |
467 default: | |
2281 | 468 assert(0); //cant happen |
2522
e25782262d7d
kill warnings patch by (M«©ns Rullg«©rd <mru inprovide com>)
michael
parents:
2398
diff
changeset
|
469 code = 0; /* kill warning */ |
0 | 470 } |
2967 | 471 |
0 | 472 #if 0 |
2967 | 473 printf("%d: %2d %2d %2d %d %d -> %d\n", j, |
0 | 474 sf[0], sf[1], sf[2], d1, d2, code); |
475 #endif | |
476 scale_code[j] = code; | |
477 sf += 3; | |
478 } | |
479 } | |
480 | |
481 /* The most important function : psycho acoustic module. In this | |
482 encoder there is basically none, so this is the worst you can do, | |
483 but also this is the simpler. */ | |
484 static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT]) | |
485 { | |
486 int i; | |
487 | |
488 for(i=0;i<s->sblimit;i++) { | |
489 smr[i] = (int)(fixed_smr[i] * 10); | |
490 } | |
491 } | |
492 | |
493 | |
494 #define SB_NOTALLOCATED 0 | |
495 #define SB_ALLOCATED 1 | |
496 #define SB_NOMORE 2 | |
497 | |
498 /* Try to maximize the smr while using a number of bits inferior to | |
499 the frame size. I tried to make the code simpler, faster and | |
500 smaller than other encoders :-) */ | |
2967 | 501 static void compute_bit_allocation(MpegAudioContext *s, |
0 | 502 short smr1[MPA_MAX_CHANNELS][SBLIMIT], |
503 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], | |
504 int *padding) | |
505 { | |
506 int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size; | |
507 int incr; | |
508 short smr[MPA_MAX_CHANNELS][SBLIMIT]; | |
509 unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT]; | |
510 const unsigned char *alloc; | |
511 | |
512 memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT); | |
513 memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT); | |
514 memset(bit_alloc, 0, s->nb_channels * SBLIMIT); | |
2967 | 515 |
0 | 516 /* compute frame size and padding */ |
517 max_frame_size = s->frame_size; | |
518 s->frame_frac += s->frame_frac_incr; | |
519 if (s->frame_frac >= 65536) { | |
520 s->frame_frac -= 65536; | |
521 s->do_padding = 1; | |
522 max_frame_size += 8; | |
523 } else { | |
524 s->do_padding = 0; | |
525 } | |
526 | |
527 /* compute the header + bit alloc size */ | |
528 current_frame_size = 32; | |
529 alloc = s->alloc_table; | |
530 for(i=0;i<s->sblimit;i++) { | |
531 incr = alloc[0]; | |
532 current_frame_size += incr * s->nb_channels; | |
533 alloc += 1 << incr; | |
534 } | |
535 for(;;) { | |
536 /* look for the subband with the largest signal to mask ratio */ | |
537 max_sb = -1; | |
538 max_ch = -1; | |
539 max_smr = 0x80000000; | |
540 for(ch=0;ch<s->nb_channels;ch++) { | |
541 for(i=0;i<s->sblimit;i++) { | |
542 if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) { | |
543 max_smr = smr[ch][i]; | |
544 max_sb = i; | |
545 max_ch = ch; | |
546 } | |
547 } | |
548 } | |
549 #if 0 | |
2967 | 550 printf("current=%d max=%d max_sb=%d alloc=%d\n", |
0 | 551 current_frame_size, max_frame_size, max_sb, |
552 bit_alloc[max_sb]); | |
2967 | 553 #endif |
0 | 554 if (max_sb < 0) |
555 break; | |
2967 | 556 |
0 | 557 /* find alloc table entry (XXX: not optimal, should use |
558 pointer table) */ | |
559 alloc = s->alloc_table; | |
560 for(i=0;i<max_sb;i++) { | |
561 alloc += 1 << alloc[0]; | |
562 } | |
563 | |
564 if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) { | |
565 /* nothing was coded for this band: add the necessary bits */ | |
566 incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6; | |
567 incr += total_quant_bits[alloc[1]]; | |
568 } else { | |
569 /* increments bit allocation */ | |
570 b = bit_alloc[max_ch][max_sb]; | |
2967 | 571 incr = total_quant_bits[alloc[b + 1]] - |
0 | 572 total_quant_bits[alloc[b]]; |
573 } | |
574 | |
575 if (current_frame_size + incr <= max_frame_size) { | |
576 /* can increase size */ | |
577 b = ++bit_alloc[max_ch][max_sb]; | |
578 current_frame_size += incr; | |
579 /* decrease smr by the resolution we added */ | |
580 smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]]; | |
581 /* max allocation size reached ? */ | |
582 if (b == ((1 << alloc[0]) - 1)) | |
583 subband_status[max_ch][max_sb] = SB_NOMORE; | |
584 else | |
585 subband_status[max_ch][max_sb] = SB_ALLOCATED; | |
586 } else { | |
587 /* cannot increase the size of this subband */ | |
588 subband_status[max_ch][max_sb] = SB_NOMORE; | |
589 } | |
590 } | |
591 *padding = max_frame_size - current_frame_size; | |
592 assert(*padding >= 0); | |
593 | |
594 #if 0 | |
595 for(i=0;i<s->sblimit;i++) { | |
596 printf("%d ", bit_alloc[i]); | |
597 } | |
598 printf("\n"); | |
599 #endif | |
600 } | |
601 | |
602 /* | |
603 * Output the mpeg audio layer 2 frame. Note how the code is small | |
604 * compared to other encoders :-) | |
605 */ | |
606 static void encode_frame(MpegAudioContext *s, | |
607 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], | |
608 int padding) | |
609 { | |
610 int i, j, k, l, bit_alloc_bits, b, ch; | |
611 unsigned char *sf; | |
612 int q[3]; | |
613 PutBitContext *p = &s->pb; | |
614 | |
615 /* header */ | |
616 | |
617 put_bits(p, 12, 0xfff); | |
618 put_bits(p, 1, 1 - s->lsf); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */ | |
619 put_bits(p, 2, 4-2); /* layer 2 */ | |
620 put_bits(p, 1, 1); /* no error protection */ | |
621 put_bits(p, 4, s->bitrate_index); | |
622 put_bits(p, 2, s->freq_index); | |
623 put_bits(p, 1, s->do_padding); /* use padding */ | |
624 put_bits(p, 1, 0); /* private_bit */ | |
625 put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO); | |
626 put_bits(p, 2, 0); /* mode_ext */ | |
627 put_bits(p, 1, 0); /* no copyright */ | |
628 put_bits(p, 1, 1); /* original */ | |
629 put_bits(p, 2, 0); /* no emphasis */ | |
630 | |
631 /* bit allocation */ | |
632 j = 0; | |
633 for(i=0;i<s->sblimit;i++) { | |
634 bit_alloc_bits = s->alloc_table[j]; | |
635 for(ch=0;ch<s->nb_channels;ch++) { | |
636 put_bits(p, bit_alloc_bits, bit_alloc[ch][i]); | |
637 } | |
638 j += 1 << bit_alloc_bits; | |
639 } | |
2967 | 640 |
0 | 641 /* scale codes */ |
642 for(i=0;i<s->sblimit;i++) { | |
643 for(ch=0;ch<s->nb_channels;ch++) { | |
2967 | 644 if (bit_alloc[ch][i]) |
0 | 645 put_bits(p, 2, s->scale_code[ch][i]); |
646 } | |
647 } | |
648 | |
649 /* scale factors */ | |
650 for(i=0;i<s->sblimit;i++) { | |
651 for(ch=0;ch<s->nb_channels;ch++) { | |
652 if (bit_alloc[ch][i]) { | |
653 sf = &s->scale_factors[ch][i][0]; | |
654 switch(s->scale_code[ch][i]) { | |
655 case 0: | |
656 put_bits(p, 6, sf[0]); | |
657 put_bits(p, 6, sf[1]); | |
658 put_bits(p, 6, sf[2]); | |
659 break; | |
660 case 3: | |
661 case 1: | |
662 put_bits(p, 6, sf[0]); | |
663 put_bits(p, 6, sf[2]); | |
664 break; | |
665 case 2: | |
666 put_bits(p, 6, sf[0]); | |
667 break; | |
668 } | |
669 } | |
670 } | |
671 } | |
2967 | 672 |
0 | 673 /* quantization & write sub band samples */ |
674 | |
675 for(k=0;k<3;k++) { | |
676 for(l=0;l<12;l+=3) { | |
677 j = 0; | |
678 for(i=0;i<s->sblimit;i++) { | |
679 bit_alloc_bits = s->alloc_table[j]; | |
680 for(ch=0;ch<s->nb_channels;ch++) { | |
681 b = bit_alloc[ch][i]; | |
682 if (b) { | |
683 int qindex, steps, m, sample, bits; | |
684 /* we encode 3 sub band samples of the same sub band at a time */ | |
685 qindex = s->alloc_table[j+b]; | |
686 steps = quant_steps[qindex]; | |
687 for(m=0;m<3;m++) { | |
688 sample = s->sb_samples[ch][k][l + m][i]; | |
689 /* divide by scale factor */ | |
690 #ifdef USE_FLOATS | |
691 { | |
692 float a; | |
693 a = (float)sample * scale_factor_inv_table[s->scale_factors[ch][i][k]]; | |
694 q[m] = (int)((a + 1.0) * steps * 0.5); | |
695 } | |
696 #else | |
697 { | |
698 int q1, e, shift, mult; | |
699 e = s->scale_factors[ch][i][k]; | |
700 shift = scale_factor_shift[e]; | |
701 mult = scale_factor_mult[e]; | |
2967 | 702 |
0 | 703 /* normalize to P bits */ |
704 if (shift < 0) | |
705 q1 = sample << (-shift); | |
706 else | |
707 q1 = sample >> shift; | |
708 q1 = (q1 * mult) >> P; | |
709 q[m] = ((q1 + (1 << P)) * steps) >> (P + 1); | |
710 } | |
711 #endif | |
712 if (q[m] >= steps) | |
713 q[m] = steps - 1; | |
714 assert(q[m] >= 0 && q[m] < steps); | |
715 } | |
716 bits = quant_bits[qindex]; | |
717 if (bits < 0) { | |
718 /* group the 3 values to save bits */ | |
2967 | 719 put_bits(p, -bits, |
0 | 720 q[0] + steps * (q[1] + steps * q[2])); |
721 #if 0 | |
2967 | 722 printf("%d: gr1 %d\n", |
0 | 723 i, q[0] + steps * (q[1] + steps * q[2])); |
724 #endif | |
725 } else { | |
726 #if 0 | |
2967 | 727 printf("%d: gr3 %d %d %d\n", |
0 | 728 i, q[0], q[1], q[2]); |
2967 | 729 #endif |
0 | 730 put_bits(p, bits, q[0]); |
731 put_bits(p, bits, q[1]); | |
732 put_bits(p, bits, q[2]); | |
733 } | |
734 } | |
735 } | |
736 /* next subband in alloc table */ | |
2967 | 737 j += 1 << bit_alloc_bits; |
0 | 738 } |
739 } | |
740 } | |
741 | |
742 /* padding */ | |
743 for(i=0;i<padding;i++) | |
744 put_bits(p, 1, 0); | |
745 | |
746 /* flush */ | |
747 flush_put_bits(p); | |
748 } | |
749 | |
1057 | 750 static int MPA_encode_frame(AVCodecContext *avctx, |
2979 | 751 unsigned char *frame, int buf_size, void *data) |
0 | 752 { |
753 MpegAudioContext *s = avctx->priv_data; | |
754 short *samples = data; | |
755 short smr[MPA_MAX_CHANNELS][SBLIMIT]; | |
756 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT]; | |
757 int padding, i; | |
758 | |
759 for(i=0;i<s->nb_channels;i++) { | |
760 filter(s, i, samples + i, s->nb_channels); | |
761 } | |
762 | |
763 for(i=0;i<s->nb_channels;i++) { | |
2967 | 764 compute_scale_factors(s->scale_code[i], s->scale_factors[i], |
0 | 765 s->sb_samples[i], s->sblimit); |
766 } | |
767 for(i=0;i<s->nb_channels;i++) { | |
768 psycho_acoustic_model(s, smr[i]); | |
769 } | |
770 compute_bit_allocation(s, smr, bit_alloc, &padding); | |
771 | |
1522
79dddc5cd990
removed the obsolete and unused parameters of init_put_bits
alex
parents:
1106
diff
changeset
|
772 init_put_bits(&s->pb, frame, MPA_MAX_CODED_FRAME_SIZE); |
0 | 773 |
774 encode_frame(s, bit_alloc, padding); | |
2967 | 775 |
0 | 776 s->nb_samples += MPA_FRAME_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:
89
diff
changeset
|
777 return pbBufPtr(&s->pb) - s->pb.buf; |
0 | 778 } |
779 | |
925 | 780 static int MPA_encode_close(AVCodecContext *avctx) |
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 |
2661
b2846918585c
a few #ifdef CONFIG_X_ENCODER, patch by (Roine Gustafsson <roine users.sourceforge net]
michael
parents:
2522
diff
changeset
|
786 #ifdef CONFIG_MP2_ENCODER |
0 | 787 AVCodec mp2_encoder = { |
788 "mp2", | |
789 CODEC_TYPE_AUDIO, | |
790 CODEC_ID_MP2, | |
791 sizeof(MpegAudioContext), | |
792 MPA_encode_init, | |
793 MPA_encode_frame, | |
925 | 794 MPA_encode_close, |
0 | 795 NULL, |
796 }; | |
2661
b2846918585c
a few #ifdef CONFIG_X_ENCODER, patch by (Roine Gustafsson <roine users.sourceforge net]
michael
parents:
2522
diff
changeset
|
797 #endif // CONFIG_MP2_ENCODER |
440
000aeeac27a2
* started to cleanup name clashes for onetime compilation
kabi
parents:
429
diff
changeset
|
798 |
000aeeac27a2
* started to cleanup name clashes for onetime compilation
kabi
parents:
429
diff
changeset
|
799 #undef FIX |