blob: 93363fe1d25f29f3332178ae1981e16105c10b3c
1 | /* |
2 | * The simplest mpeg audio layer 2 encoder |
3 | * Copyright (c) 2000, 2001 Fabrice Bellard |
4 | * |
5 | * This file is part of FFmpeg. |
6 | * |
7 | * FFmpeg is free software; you can redistribute it and/or |
8 | * modify it under the terms of the GNU Lesser General Public |
9 | * License as published by the Free Software Foundation; either |
10 | * version 2.1 of the License, or (at your option) any later version. |
11 | * |
12 | * FFmpeg is distributed in the hope that it will be useful, |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
15 | * Lesser General Public License for more details. |
16 | * |
17 | * You should have received a copy of the GNU Lesser General Public |
18 | * License along with FFmpeg; if not, write to the Free Software |
19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
20 | */ |
21 | |
22 | /** |
23 | * @file |
24 | * The simplest mpeg audio layer 2 encoder. |
25 | */ |
26 | |
27 | #include "libavutil/channel_layout.h" |
28 | |
29 | #include "avcodec.h" |
30 | #include "internal.h" |
31 | #include "put_bits.h" |
32 | |
33 | #define FRAC_BITS 15 /* fractional bits for sb_samples and dct */ |
34 | #define WFRAC_BITS 14 /* fractional bits for window */ |
35 | |
36 | #include "mpegaudio.h" |
37 | #include "mpegaudiodsp.h" |
38 | #include "mpegaudiodata.h" |
39 | #include "mpegaudiotab.h" |
40 | |
41 | /* currently, cannot change these constants (need to modify |
42 | quantization stage) */ |
43 | #define MUL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS) |
44 | |
45 | #define SAMPLES_BUF_SIZE 4096 |
46 | |
47 | typedef struct MpegAudioContext { |
48 | PutBitContext pb; |
49 | int nb_channels; |
50 | int lsf; /* 1 if mpeg2 low bitrate selected */ |
51 | int bitrate_index; /* bit rate */ |
52 | int freq_index; |
53 | int frame_size; /* frame size, in bits, without padding */ |
54 | /* padding computation */ |
55 | int frame_frac, frame_frac_incr, do_padding; |
56 | short samples_buf[MPA_MAX_CHANNELS][SAMPLES_BUF_SIZE]; /* buffer for filter */ |
57 | int samples_offset[MPA_MAX_CHANNELS]; /* offset in samples_buf */ |
58 | int sb_samples[MPA_MAX_CHANNELS][3][12][SBLIMIT]; |
59 | unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3]; /* scale factors */ |
60 | /* code to group 3 scale factors */ |
61 | unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT]; |
62 | int sblimit; /* number of used subbands */ |
63 | const unsigned char *alloc_table; |
64 | int16_t filter_bank[512]; |
65 | int scale_factor_table[64]; |
66 | unsigned char scale_diff_table[128]; |
67 | #if USE_FLOATS |
68 | float scale_factor_inv_table[64]; |
69 | #else |
70 | int8_t scale_factor_shift[64]; |
71 | unsigned short scale_factor_mult[64]; |
72 | #endif |
73 | unsigned short total_quant_bits[17]; /* total number of bits per allocation group */ |
74 | } MpegAudioContext; |
75 | |
76 | static av_cold int MPA_encode_init(AVCodecContext *avctx) |
77 | { |
78 | MpegAudioContext *s = avctx->priv_data; |
79 | int freq = avctx->sample_rate; |
80 | int bitrate = avctx->bit_rate; |
81 | int channels = avctx->channels; |
82 | int i, v, table; |
83 | float a; |
84 | |
85 | if (channels <= 0 || channels > 2){ |
86 | av_log(avctx, AV_LOG_ERROR, "encoding %d channel(s) is not allowed in mp2\n", channels); |
87 | return AVERROR(EINVAL); |
88 | } |
89 | bitrate = bitrate / 1000; |
90 | s->nb_channels = channels; |
91 | avctx->frame_size = MPA_FRAME_SIZE; |
92 | avctx->initial_padding = 512 - 32 + 1; |
93 | |
94 | /* encoding freq */ |
95 | s->lsf = 0; |
96 | for(i=0;i<3;i++) { |
97 | if (avpriv_mpa_freq_tab[i] == freq) |
98 | break; |
99 | if ((avpriv_mpa_freq_tab[i] / 2) == freq) { |
100 | s->lsf = 1; |
101 | break; |
102 | } |
103 | } |
104 | if (i == 3){ |
105 | av_log(avctx, AV_LOG_ERROR, "Sampling rate %d is not allowed in mp2\n", freq); |
106 | return AVERROR(EINVAL); |
107 | } |
108 | s->freq_index = i; |
109 | |
110 | /* encoding bitrate & frequency */ |
111 | for(i=1;i<15;i++) { |
112 | if (avpriv_mpa_bitrate_tab[s->lsf][1][i] == bitrate) |
113 | break; |
114 | } |
115 | if (i == 15 && !avctx->bit_rate) { |
116 | i = 14; |
117 | bitrate = avpriv_mpa_bitrate_tab[s->lsf][1][i]; |
118 | avctx->bit_rate = bitrate * 1000; |
119 | } |
120 | if (i == 15){ |
121 | av_log(avctx, AV_LOG_ERROR, "bitrate %d is not allowed in mp2\n", bitrate); |
122 | return AVERROR(EINVAL); |
123 | } |
124 | s->bitrate_index = i; |
125 | |
126 | /* compute total header size & pad bit */ |
127 | |
128 | a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0); |
129 | s->frame_size = ((int)a) * 8; |
130 | |
131 | /* frame fractional size to compute padding */ |
132 | s->frame_frac = 0; |
133 | s->frame_frac_incr = (int)((a - floor(a)) * 65536.0); |
134 | |
135 | /* select the right allocation table */ |
136 | table = ff_mpa_l2_select_table(bitrate, s->nb_channels, freq, s->lsf); |
137 | |
138 | /* number of used subbands */ |
139 | s->sblimit = ff_mpa_sblimit_table[table]; |
140 | s->alloc_table = ff_mpa_alloc_tables[table]; |
141 | |
142 | ff_dlog(avctx, "%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n", |
143 | bitrate, freq, s->frame_size, table, s->frame_frac_incr); |
144 | |
145 | for(i=0;i<s->nb_channels;i++) |
146 | s->samples_offset[i] = 0; |
147 | |
148 | for(i=0;i<257;i++) { |
149 | int v; |
150 | v = ff_mpa_enwindow[i]; |
151 | #if WFRAC_BITS != 16 |
152 | v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS); |
153 | #endif |
154 | s->filter_bank[i] = v; |
155 | if ((i & 63) != 0) |
156 | v = -v; |
157 | if (i != 0) |
158 | s->filter_bank[512 - i] = v; |
159 | } |
160 | |
161 | for(i=0;i<64;i++) { |
162 | v = (int)(exp2((3 - i) / 3.0) * (1 << 20)); |
163 | if (v <= 0) |
164 | v = 1; |
165 | s->scale_factor_table[i] = v; |
166 | #if USE_FLOATS |
167 | s->scale_factor_inv_table[i] = exp2(-(3 - i) / 3.0) / (float)(1 << 20); |
168 | #else |
169 | #define P 15 |
170 | s->scale_factor_shift[i] = 21 - P - (i / 3); |
171 | s->scale_factor_mult[i] = (1 << P) * exp2((i % 3) / 3.0); |
172 | #endif |
173 | } |
174 | for(i=0;i<128;i++) { |
175 | v = i - 64; |
176 | if (v <= -3) |
177 | v = 0; |
178 | else if (v < 0) |
179 | v = 1; |
180 | else if (v == 0) |
181 | v = 2; |
182 | else if (v < 3) |
183 | v = 3; |
184 | else |
185 | v = 4; |
186 | s->scale_diff_table[i] = v; |
187 | } |
188 | |
189 | for(i=0;i<17;i++) { |
190 | v = ff_mpa_quant_bits[i]; |
191 | if (v < 0) |
192 | v = -v; |
193 | else |
194 | v = v * 3; |
195 | s->total_quant_bits[i] = 12 * v; |
196 | } |
197 | |
198 | return 0; |
199 | } |
200 | |
201 | /* 32 point floating point IDCT without 1/sqrt(2) coef zero scaling */ |
202 | static void idct32(int *out, int *tab) |
203 | { |
204 | int i, j; |
205 | int *t, *t1, xr; |
206 | const int *xp = costab32; |
207 | |
208 | for(j=31;j>=3;j-=2) tab[j] += tab[j - 2]; |
209 | |
210 | t = tab + 30; |
211 | t1 = tab + 2; |
212 | do { |
213 | t[0] += t[-4]; |
214 | t[1] += t[1 - 4]; |
215 | t -= 4; |
216 | } while (t != t1); |
217 | |
218 | t = tab + 28; |
219 | t1 = tab + 4; |
220 | do { |
221 | t[0] += t[-8]; |
222 | t[1] += t[1-8]; |
223 | t[2] += t[2-8]; |
224 | t[3] += t[3-8]; |
225 | t -= 8; |
226 | } while (t != t1); |
227 | |
228 | t = tab; |
229 | t1 = tab + 32; |
230 | do { |
231 | t[ 3] = -t[ 3]; |
232 | t[ 6] = -t[ 6]; |
233 | |
234 | t[11] = -t[11]; |
235 | t[12] = -t[12]; |
236 | t[13] = -t[13]; |
237 | t[15] = -t[15]; |
238 | t += 16; |
239 | } while (t != t1); |
240 | |
241 | |
242 | t = tab; |
243 | t1 = tab + 8; |
244 | do { |
245 | int x1, x2, x3, x4; |
246 | |
247 | x3 = MUL(t[16], FIX(M_SQRT2*0.5)); |
248 | x4 = t[0] - x3; |
249 | x3 = t[0] + x3; |
250 | |
251 | x2 = MUL(-(t[24] + t[8]), FIX(M_SQRT2*0.5)); |
252 | x1 = MUL((t[8] - x2), xp[0]); |
253 | x2 = MUL((t[8] + x2), xp[1]); |
254 | |
255 | t[ 0] = x3 + x1; |
256 | t[ 8] = x4 - x2; |
257 | t[16] = x4 + x2; |
258 | t[24] = x3 - x1; |
259 | t++; |
260 | } while (t != t1); |
261 | |
262 | xp += 2; |
263 | t = tab; |
264 | t1 = tab + 4; |
265 | do { |
266 | xr = MUL(t[28],xp[0]); |
267 | t[28] = (t[0] - xr); |
268 | t[0] = (t[0] + xr); |
269 | |
270 | xr = MUL(t[4],xp[1]); |
271 | t[ 4] = (t[24] - xr); |
272 | t[24] = (t[24] + xr); |
273 | |
274 | xr = MUL(t[20],xp[2]); |
275 | t[20] = (t[8] - xr); |
276 | t[ 8] = (t[8] + xr); |
277 | |
278 | xr = MUL(t[12],xp[3]); |
279 | t[12] = (t[16] - xr); |
280 | t[16] = (t[16] + xr); |
281 | t++; |
282 | } while (t != t1); |
283 | xp += 4; |
284 | |
285 | for (i = 0; i < 4; i++) { |
286 | xr = MUL(tab[30-i*4],xp[0]); |
287 | tab[30-i*4] = (tab[i*4] - xr); |
288 | tab[ i*4] = (tab[i*4] + xr); |
289 | |
290 | xr = MUL(tab[ 2+i*4],xp[1]); |
291 | tab[ 2+i*4] = (tab[28-i*4] - xr); |
292 | tab[28-i*4] = (tab[28-i*4] + xr); |
293 | |
294 | xr = MUL(tab[31-i*4],xp[0]); |
295 | tab[31-i*4] = (tab[1+i*4] - xr); |
296 | tab[ 1+i*4] = (tab[1+i*4] + xr); |
297 | |
298 | xr = MUL(tab[ 3+i*4],xp[1]); |
299 | tab[ 3+i*4] = (tab[29-i*4] - xr); |
300 | tab[29-i*4] = (tab[29-i*4] + xr); |
301 | |
302 | xp += 2; |
303 | } |
304 | |
305 | t = tab + 30; |
306 | t1 = tab + 1; |
307 | do { |
308 | xr = MUL(t1[0], *xp); |
309 | t1[0] = (t[0] - xr); |
310 | t[0] = (t[0] + xr); |
311 | t -= 2; |
312 | t1 += 2; |
313 | xp++; |
314 | } while (t >= tab); |
315 | |
316 | for(i=0;i<32;i++) { |
317 | out[i] = tab[bitinv32[i]]; |
318 | } |
319 | } |
320 | |
321 | #define WSHIFT (WFRAC_BITS + 15 - FRAC_BITS) |
322 | |
323 | static void filter(MpegAudioContext *s, int ch, const short *samples, int incr) |
324 | { |
325 | short *p, *q; |
326 | int sum, offset, i, j; |
327 | int tmp[64]; |
328 | int tmp1[32]; |
329 | int *out; |
330 | |
331 | offset = s->samples_offset[ch]; |
332 | out = &s->sb_samples[ch][0][0][0]; |
333 | for(j=0;j<36;j++) { |
334 | /* 32 samples at once */ |
335 | for(i=0;i<32;i++) { |
336 | s->samples_buf[ch][offset + (31 - i)] = samples[0]; |
337 | samples += incr; |
338 | } |
339 | |
340 | /* filter */ |
341 | p = s->samples_buf[ch] + offset; |
342 | q = s->filter_bank; |
343 | /* maxsum = 23169 */ |
344 | for(i=0;i<64;i++) { |
345 | sum = p[0*64] * q[0*64]; |
346 | sum += p[1*64] * q[1*64]; |
347 | sum += p[2*64] * q[2*64]; |
348 | sum += p[3*64] * q[3*64]; |
349 | sum += p[4*64] * q[4*64]; |
350 | sum += p[5*64] * q[5*64]; |
351 | sum += p[6*64] * q[6*64]; |
352 | sum += p[7*64] * q[7*64]; |
353 | tmp[i] = sum; |
354 | p++; |
355 | q++; |
356 | } |
357 | tmp1[0] = tmp[16] >> WSHIFT; |
358 | for( i=1; i<=16; i++ ) tmp1[i] = (tmp[i+16]+tmp[16-i]) >> WSHIFT; |
359 | for( i=17; i<=31; i++ ) tmp1[i] = (tmp[i+16]-tmp[80-i]) >> WSHIFT; |
360 | |
361 | idct32(out, tmp1); |
362 | |
363 | /* advance of 32 samples */ |
364 | offset -= 32; |
365 | out += 32; |
366 | /* handle the wrap around */ |
367 | if (offset < 0) { |
368 | memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32), |
369 | s->samples_buf[ch], (512 - 32) * 2); |
370 | offset = SAMPLES_BUF_SIZE - 512; |
371 | } |
372 | } |
373 | s->samples_offset[ch] = offset; |
374 | } |
375 | |
376 | static void compute_scale_factors(MpegAudioContext *s, |
377 | unsigned char scale_code[SBLIMIT], |
378 | unsigned char scale_factors[SBLIMIT][3], |
379 | int sb_samples[3][12][SBLIMIT], |
380 | int sblimit) |
381 | { |
382 | int *p, vmax, v, n, i, j, k, code; |
383 | int index, d1, d2; |
384 | unsigned char *sf = &scale_factors[0][0]; |
385 | |
386 | for(j=0;j<sblimit;j++) { |
387 | for(i=0;i<3;i++) { |
388 | /* find the max absolute value */ |
389 | p = &sb_samples[i][0][j]; |
390 | vmax = abs(*p); |
391 | for(k=1;k<12;k++) { |
392 | p += SBLIMIT; |
393 | v = abs(*p); |
394 | if (v > vmax) |
395 | vmax = v; |
396 | } |
397 | /* compute the scale factor index using log 2 computations */ |
398 | if (vmax > 1) { |
399 | n = av_log2(vmax); |
400 | /* n is the position of the MSB of vmax. now |
401 | use at most 2 compares to find the index */ |
402 | index = (21 - n) * 3 - 3; |
403 | if (index >= 0) { |
404 | while (vmax <= s->scale_factor_table[index+1]) |
405 | index++; |
406 | } else { |
407 | index = 0; /* very unlikely case of overflow */ |
408 | } |
409 | } else { |
410 | index = 62; /* value 63 is not allowed */ |
411 | } |
412 | |
413 | ff_dlog(NULL, "%2d:%d in=%x %x %d\n", |
414 | j, i, vmax, s->scale_factor_table[index], index); |
415 | /* store the scale factor */ |
416 | av_assert2(index >=0 && index <= 63); |
417 | sf[i] = index; |
418 | } |
419 | |
420 | /* compute the transmission factor : look if the scale factors |
421 | are close enough to each other */ |
422 | d1 = s->scale_diff_table[sf[0] - sf[1] + 64]; |
423 | d2 = s->scale_diff_table[sf[1] - sf[2] + 64]; |
424 | |
425 | /* handle the 25 cases */ |
426 | switch(d1 * 5 + d2) { |
427 | case 0*5+0: |
428 | case 0*5+4: |
429 | case 3*5+4: |
430 | case 4*5+0: |
431 | case 4*5+4: |
432 | code = 0; |
433 | break; |
434 | case 0*5+1: |
435 | case 0*5+2: |
436 | case 4*5+1: |
437 | case 4*5+2: |
438 | code = 3; |
439 | sf[2] = sf[1]; |
440 | break; |
441 | case 0*5+3: |
442 | case 4*5+3: |
443 | code = 3; |
444 | sf[1] = sf[2]; |
445 | break; |
446 | case 1*5+0: |
447 | case 1*5+4: |
448 | case 2*5+4: |
449 | code = 1; |
450 | sf[1] = sf[0]; |
451 | break; |
452 | case 1*5+1: |
453 | case 1*5+2: |
454 | case 2*5+0: |
455 | case 2*5+1: |
456 | case 2*5+2: |
457 | code = 2; |
458 | sf[1] = sf[2] = sf[0]; |
459 | break; |
460 | case 2*5+3: |
461 | case 3*5+3: |
462 | code = 2; |
463 | sf[0] = sf[1] = sf[2]; |
464 | break; |
465 | case 3*5+0: |
466 | case 3*5+1: |
467 | case 3*5+2: |
468 | code = 2; |
469 | sf[0] = sf[2] = sf[1]; |
470 | break; |
471 | case 1*5+3: |
472 | code = 2; |
473 | if (sf[0] > sf[2]) |
474 | sf[0] = sf[2]; |
475 | sf[1] = sf[2] = sf[0]; |
476 | break; |
477 | default: |
478 | av_assert2(0); //cannot happen |
479 | code = 0; /* kill warning */ |
480 | } |
481 | |
482 | ff_dlog(NULL, "%d: %2d %2d %2d %d %d -> %d\n", j, |
483 | sf[0], sf[1], sf[2], d1, d2, code); |
484 | scale_code[j] = code; |
485 | sf += 3; |
486 | } |
487 | } |
488 | |
489 | /* The most important function : psycho acoustic module. In this |
490 | encoder there is basically none, so this is the worst you can do, |
491 | but also this is the simpler. */ |
492 | static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT]) |
493 | { |
494 | int i; |
495 | |
496 | for(i=0;i<s->sblimit;i++) { |
497 | smr[i] = (int)(fixed_smr[i] * 10); |
498 | } |
499 | } |
500 | |
501 | |
502 | #define SB_NOTALLOCATED 0 |
503 | #define SB_ALLOCATED 1 |
504 | #define SB_NOMORE 2 |
505 | |
506 | /* Try to maximize the smr while using a number of bits inferior to |
507 | the frame size. I tried to make the code simpler, faster and |
508 | smaller than other encoders :-) */ |
509 | static void compute_bit_allocation(MpegAudioContext *s, |
510 | short smr1[MPA_MAX_CHANNELS][SBLIMIT], |
511 | unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], |
512 | int *padding) |
513 | { |
514 | int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size; |
515 | int incr; |
516 | short smr[MPA_MAX_CHANNELS][SBLIMIT]; |
517 | unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT]; |
518 | const unsigned char *alloc; |
519 | |
520 | memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT); |
521 | memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT); |
522 | memset(bit_alloc, 0, s->nb_channels * SBLIMIT); |
523 | |
524 | /* compute frame size and padding */ |
525 | max_frame_size = s->frame_size; |
526 | s->frame_frac += s->frame_frac_incr; |
527 | if (s->frame_frac >= 65536) { |
528 | s->frame_frac -= 65536; |
529 | s->do_padding = 1; |
530 | max_frame_size += 8; |
531 | } else { |
532 | s->do_padding = 0; |
533 | } |
534 | |
535 | /* compute the header + bit alloc size */ |
536 | current_frame_size = 32; |
537 | alloc = s->alloc_table; |
538 | for(i=0;i<s->sblimit;i++) { |
539 | incr = alloc[0]; |
540 | current_frame_size += incr * s->nb_channels; |
541 | alloc += 1 << incr; |
542 | } |
543 | for(;;) { |
544 | /* look for the subband with the largest signal to mask ratio */ |
545 | max_sb = -1; |
546 | max_ch = -1; |
547 | max_smr = INT_MIN; |
548 | for(ch=0;ch<s->nb_channels;ch++) { |
549 | for(i=0;i<s->sblimit;i++) { |
550 | if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) { |
551 | max_smr = smr[ch][i]; |
552 | max_sb = i; |
553 | max_ch = ch; |
554 | } |
555 | } |
556 | } |
557 | if (max_sb < 0) |
558 | break; |
559 | ff_dlog(NULL, "current=%d max=%d max_sb=%d max_ch=%d alloc=%d\n", |
560 | current_frame_size, max_frame_size, max_sb, max_ch, |
561 | bit_alloc[max_ch][max_sb]); |
562 | |
563 | /* find alloc table entry (XXX: not optimal, should use |
564 | pointer table) */ |
565 | alloc = s->alloc_table; |
566 | for(i=0;i<max_sb;i++) { |
567 | alloc += 1 << alloc[0]; |
568 | } |
569 | |
570 | if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) { |
571 | /* nothing was coded for this band: add the necessary bits */ |
572 | incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6; |
573 | incr += s->total_quant_bits[alloc[1]]; |
574 | } else { |
575 | /* increments bit allocation */ |
576 | b = bit_alloc[max_ch][max_sb]; |
577 | incr = s->total_quant_bits[alloc[b + 1]] - |
578 | s->total_quant_bits[alloc[b]]; |
579 | } |
580 | |
581 | if (current_frame_size + incr <= max_frame_size) { |
582 | /* can increase size */ |
583 | b = ++bit_alloc[max_ch][max_sb]; |
584 | current_frame_size += incr; |
585 | /* decrease smr by the resolution we added */ |
586 | smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]]; |
587 | /* max allocation size reached ? */ |
588 | if (b == ((1 << alloc[0]) - 1)) |
589 | subband_status[max_ch][max_sb] = SB_NOMORE; |
590 | else |
591 | subband_status[max_ch][max_sb] = SB_ALLOCATED; |
592 | } else { |
593 | /* cannot increase the size of this subband */ |
594 | subband_status[max_ch][max_sb] = SB_NOMORE; |
595 | } |
596 | } |
597 | *padding = max_frame_size - current_frame_size; |
598 | av_assert0(*padding >= 0); |
599 | } |
600 | |
601 | /* |
602 | * Output the MPEG audio layer 2 frame. Note how the code is small |
603 | * compared to other encoders :-) |
604 | */ |
605 | static void encode_frame(MpegAudioContext *s, |
606 | unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], |
607 | int padding) |
608 | { |
609 | int i, j, k, l, bit_alloc_bits, b, ch; |
610 | unsigned char *sf; |
611 | int q[3]; |
612 | PutBitContext *p = &s->pb; |
613 | |
614 | /* header */ |
615 | |
616 | put_bits(p, 12, 0xfff); |
617 | put_bits(p, 1, 1 - s->lsf); /* 1 = MPEG-1 ID, 0 = MPEG-2 lsf ID */ |
618 | put_bits(p, 2, 4-2); /* layer 2 */ |
619 | put_bits(p, 1, 1); /* no error protection */ |
620 | put_bits(p, 4, s->bitrate_index); |
621 | put_bits(p, 2, s->freq_index); |
622 | put_bits(p, 1, s->do_padding); /* use padding */ |
623 | put_bits(p, 1, 0); /* private_bit */ |
624 | put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO); |
625 | put_bits(p, 2, 0); /* mode_ext */ |
626 | put_bits(p, 1, 0); /* no copyright */ |
627 | put_bits(p, 1, 1); /* original */ |
628 | put_bits(p, 2, 0); /* no emphasis */ |
629 | |
630 | /* bit allocation */ |
631 | j = 0; |
632 | for(i=0;i<s->sblimit;i++) { |
633 | bit_alloc_bits = s->alloc_table[j]; |
634 | for(ch=0;ch<s->nb_channels;ch++) { |
635 | put_bits(p, bit_alloc_bits, bit_alloc[ch][i]); |
636 | } |
637 | j += 1 << bit_alloc_bits; |
638 | } |
639 | |
640 | /* scale codes */ |
641 | for(i=0;i<s->sblimit;i++) { |
642 | for(ch=0;ch<s->nb_channels;ch++) { |
643 | if (bit_alloc[ch][i]) |
644 | put_bits(p, 2, s->scale_code[ch][i]); |
645 | } |
646 | } |
647 | |
648 | /* scale factors */ |
649 | for(i=0;i<s->sblimit;i++) { |
650 | for(ch=0;ch<s->nb_channels;ch++) { |
651 | if (bit_alloc[ch][i]) { |
652 | sf = &s->scale_factors[ch][i][0]; |
653 | switch(s->scale_code[ch][i]) { |
654 | case 0: |
655 | put_bits(p, 6, sf[0]); |
656 | put_bits(p, 6, sf[1]); |
657 | put_bits(p, 6, sf[2]); |
658 | break; |
659 | case 3: |
660 | case 1: |
661 | put_bits(p, 6, sf[0]); |
662 | put_bits(p, 6, sf[2]); |
663 | break; |
664 | case 2: |
665 | put_bits(p, 6, sf[0]); |
666 | break; |
667 | } |
668 | } |
669 | } |
670 | } |
671 | |
672 | /* quantization & write sub band samples */ |
673 | |
674 | for(k=0;k<3;k++) { |
675 | for(l=0;l<12;l+=3) { |
676 | j = 0; |
677 | for(i=0;i<s->sblimit;i++) { |
678 | bit_alloc_bits = s->alloc_table[j]; |
679 | for(ch=0;ch<s->nb_channels;ch++) { |
680 | b = bit_alloc[ch][i]; |
681 | if (b) { |
682 | int qindex, steps, m, sample, bits; |
683 | /* we encode 3 sub band samples of the same sub band at a time */ |
684 | qindex = s->alloc_table[j+b]; |
685 | steps = ff_mpa_quant_steps[qindex]; |
686 | for(m=0;m<3;m++) { |
687 | sample = s->sb_samples[ch][k][l + m][i]; |
688 | /* divide by scale factor */ |
689 | #if USE_FLOATS |
690 | { |
691 | float a; |
692 | a = (float)sample * s->scale_factor_inv_table[s->scale_factors[ch][i][k]]; |
693 | q[m] = (int)((a + 1.0) * steps * 0.5); |
694 | } |
695 | #else |
696 | { |
697 | int q1, e, shift, mult; |
698 | e = s->scale_factors[ch][i][k]; |
699 | shift = s->scale_factor_shift[e]; |
700 | mult = s->scale_factor_mult[e]; |
701 | |
702 | /* normalize to P bits */ |
703 | if (shift < 0) |
704 | q1 = sample << (-shift); |
705 | else |
706 | q1 = sample >> shift; |
707 | q1 = (q1 * mult) >> P; |
708 | q1 += 1 << P; |
709 | if (q1 < 0) |
710 | q1 = 0; |
711 | q[m] = (q1 * (unsigned)steps) >> (P + 1); |
712 | } |
713 | #endif |
714 | if (q[m] >= steps) |
715 | q[m] = steps - 1; |
716 | av_assert2(q[m] >= 0 && q[m] < steps); |
717 | } |
718 | bits = ff_mpa_quant_bits[qindex]; |
719 | if (bits < 0) { |
720 | /* group the 3 values to save bits */ |
721 | put_bits(p, -bits, |
722 | q[0] + steps * (q[1] + steps * q[2])); |
723 | } else { |
724 | put_bits(p, bits, q[0]); |
725 | put_bits(p, bits, q[1]); |
726 | put_bits(p, bits, q[2]); |
727 | } |
728 | } |
729 | } |
730 | /* next subband in alloc table */ |
731 | j += 1 << bit_alloc_bits; |
732 | } |
733 | } |
734 | } |
735 | |
736 | /* padding */ |
737 | for(i=0;i<padding;i++) |
738 | put_bits(p, 1, 0); |
739 | |
740 | /* flush */ |
741 | flush_put_bits(p); |
742 | } |
743 | |
744 | static int MPA_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, |
745 | const AVFrame *frame, int *got_packet_ptr) |
746 | { |
747 | MpegAudioContext *s = avctx->priv_data; |
748 | const int16_t *samples = (const int16_t *)frame->data[0]; |
749 | short smr[MPA_MAX_CHANNELS][SBLIMIT]; |
750 | unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT]; |
751 | int padding, i, ret; |
752 | |
753 | for(i=0;i<s->nb_channels;i++) { |
754 | filter(s, i, samples + i, s->nb_channels); |
755 | } |
756 | |
757 | for(i=0;i<s->nb_channels;i++) { |
758 | compute_scale_factors(s, s->scale_code[i], s->scale_factors[i], |
759 | s->sb_samples[i], s->sblimit); |
760 | } |
761 | for(i=0;i<s->nb_channels;i++) { |
762 | psycho_acoustic_model(s, smr[i]); |
763 | } |
764 | compute_bit_allocation(s, smr, bit_alloc, &padding); |
765 | |
766 | if ((ret = ff_alloc_packet2(avctx, avpkt, MPA_MAX_CODED_FRAME_SIZE, 0)) < 0) |
767 | return ret; |
768 | |
769 | init_put_bits(&s->pb, avpkt->data, avpkt->size); |
770 | |
771 | encode_frame(s, bit_alloc, padding); |
772 | |
773 | if (frame->pts != AV_NOPTS_VALUE) |
774 | avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->initial_padding); |
775 | |
776 | avpkt->size = put_bits_count(&s->pb) / 8; |
777 | *got_packet_ptr = 1; |
778 | return 0; |
779 | } |
780 | |
781 | static const AVCodecDefault mp2_defaults[] = { |
782 | { "b", "0" }, |
783 | { NULL }, |
784 | }; |
785 | |
786 |