0
|
1 /*
|
|
2 * The simplest mpeg audio layer 2 encoder
|
|
3 * Copyright (c) 2000 Gerard Lantau.
|
|
4 *
|
|
5 * This program is free software; you can redistribute it and/or modify
|
|
6 * it under the terms of the GNU General Public License as published by
|
|
7 * the Free Software Foundation; either version 2 of the License, or
|
|
8 * (at your option) any later version.
|
|
9 *
|
|
10 * This program is distributed in the hope that it will be useful,
|
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
13 * GNU General Public License for more details.
|
|
14 *
|
|
15 * You should have received a copy of the GNU General Public License
|
|
16 * along with this program; if not, write to the Free Software
|
|
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
|
|
18 */
|
64
|
19 #include "avcodec.h"
|
0
|
20 #include <math.h>
|
|
21 #include "mpegaudio.h"
|
|
22
|
|
23 /* define it to use floats in quantization (I don't like floats !) */
|
|
24 //#define USE_FLOATS
|
|
25
|
|
26 #define MPA_STEREO 0
|
|
27 #define MPA_JSTEREO 1
|
|
28 #define MPA_DUAL 2
|
|
29 #define MPA_MONO 3
|
|
30
|
|
31 #include "mpegaudiotab.h"
|
|
32
|
|
33 int MPA_encode_init(AVCodecContext *avctx)
|
|
34 {
|
|
35 MpegAudioContext *s = avctx->priv_data;
|
|
36 int freq = avctx->sample_rate;
|
|
37 int bitrate = avctx->bit_rate;
|
|
38 int channels = avctx->channels;
|
|
39 int i, v, table, ch_bitrate;
|
|
40 float a;
|
|
41
|
|
42 if (channels > 2)
|
|
43 return -1;
|
|
44 bitrate = bitrate / 1000;
|
|
45 s->nb_channels = channels;
|
|
46 s->freq = freq;
|
|
47 s->bit_rate = bitrate * 1000;
|
|
48 avctx->frame_size = MPA_FRAME_SIZE;
|
|
49 avctx->key_frame = 1; /* always key frame */
|
|
50
|
|
51 /* encoding freq */
|
|
52 s->lsf = 0;
|
|
53 for(i=0;i<3;i++) {
|
|
54 if (freq_tab[i] == freq)
|
|
55 break;
|
|
56 if ((freq_tab[i] / 2) == freq) {
|
|
57 s->lsf = 1;
|
|
58 break;
|
|
59 }
|
|
60 }
|
|
61 if (i == 3)
|
|
62 return -1;
|
|
63 s->freq_index = i;
|
|
64
|
|
65 /* encoding bitrate & frequency */
|
|
66 for(i=0;i<15;i++) {
|
|
67 if (bitrate_tab[1-s->lsf][i] == bitrate)
|
|
68 break;
|
|
69 }
|
|
70 if (i == 15)
|
|
71 return -1;
|
|
72 s->bitrate_index = i;
|
|
73
|
|
74 /* compute total header size & pad bit */
|
|
75
|
|
76 a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0);
|
|
77 s->frame_size = ((int)a) * 8;
|
|
78
|
|
79 /* frame fractional size to compute padding */
|
|
80 s->frame_frac = 0;
|
|
81 s->frame_frac_incr = (int)((a - floor(a)) * 65536.0);
|
|
82
|
|
83 /* select the right allocation table */
|
|
84 ch_bitrate = bitrate / s->nb_channels;
|
|
85 if (!s->lsf) {
|
|
86 if ((freq == 48000 && ch_bitrate >= 56) ||
|
|
87 (ch_bitrate >= 56 && ch_bitrate <= 80))
|
|
88 table = 0;
|
|
89 else if (freq != 48000 && ch_bitrate >= 96)
|
|
90 table = 1;
|
|
91 else if (freq != 32000 && ch_bitrate <= 48)
|
|
92 table = 2;
|
|
93 else
|
|
94 table = 3;
|
|
95 } else {
|
|
96 table = 4;
|
|
97 }
|
|
98 /* number of used subbands */
|
|
99 s->sblimit = sblimit_table[table];
|
|
100 s->alloc_table = alloc_tables[table];
|
|
101
|
|
102 #ifdef DEBUG
|
|
103 printf("%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n",
|
|
104 bitrate, freq, s->frame_size, table, s->frame_frac_incr);
|
|
105 #endif
|
|
106
|
|
107 for(i=0;i<s->nb_channels;i++)
|
|
108 s->samples_offset[i] = 0;
|
|
109
|
|
110 for(i=0;i<512;i++) {
|
|
111 float a = enwindow[i] * 32768.0 * 16.0;
|
|
112 filter_bank[i] = (int)(a);
|
|
113 }
|
|
114 for(i=0;i<64;i++) {
|
|
115 v = (int)(pow(2.0, (3 - i) / 3.0) * (1 << 20));
|
|
116 if (v <= 0)
|
|
117 v = 1;
|
|
118 scale_factor_table[i] = v;
|
|
119 #ifdef USE_FLOATS
|
|
120 scale_factor_inv_table[i] = pow(2.0, -(3 - i) / 3.0) / (float)(1 << 20);
|
|
121 #else
|
|
122 #define P 15
|
|
123 scale_factor_shift[i] = 21 - P - (i / 3);
|
|
124 scale_factor_mult[i] = (1 << P) * pow(2.0, (i % 3) / 3.0);
|
|
125 #endif
|
|
126 }
|
|
127 for(i=0;i<128;i++) {
|
|
128 v = i - 64;
|
|
129 if (v <= -3)
|
|
130 v = 0;
|
|
131 else if (v < 0)
|
|
132 v = 1;
|
|
133 else if (v == 0)
|
|
134 v = 2;
|
|
135 else if (v < 3)
|
|
136 v = 3;
|
|
137 else
|
|
138 v = 4;
|
|
139 scale_diff_table[i] = v;
|
|
140 }
|
|
141
|
|
142 for(i=0;i<17;i++) {
|
|
143 v = quant_bits[i];
|
|
144 if (v < 0)
|
|
145 v = -v;
|
|
146 else
|
|
147 v = v * 3;
|
|
148 total_quant_bits[i] = 12 * v;
|
|
149 }
|
|
150
|
|
151 return 0;
|
|
152 }
|
|
153
|
|
154 /* 32 point floating point IDCT */
|
|
155 static void idct32(int *out, int *tab, int sblimit, int left_shift)
|
|
156 {
|
|
157 int i, j;
|
|
158 int *t, *t1, xr;
|
|
159 const int *xp = costab32;
|
|
160
|
|
161 for(j=31;j>=3;j-=2) tab[j] += tab[j - 2];
|
|
162
|
|
163 t = tab + 30;
|
|
164 t1 = tab + 2;
|
|
165 do {
|
|
166 t[0] += t[-4];
|
|
167 t[1] += t[1 - 4];
|
|
168 t -= 4;
|
|
169 } while (t != t1);
|
|
170
|
|
171 t = tab + 28;
|
|
172 t1 = tab + 4;
|
|
173 do {
|
|
174 t[0] += t[-8];
|
|
175 t[1] += t[1-8];
|
|
176 t[2] += t[2-8];
|
|
177 t[3] += t[3-8];
|
|
178 t -= 8;
|
|
179 } while (t != t1);
|
|
180
|
|
181 t = tab;
|
|
182 t1 = tab + 32;
|
|
183 do {
|
|
184 t[ 3] = -t[ 3];
|
|
185 t[ 6] = -t[ 6];
|
|
186
|
|
187 t[11] = -t[11];
|
|
188 t[12] = -t[12];
|
|
189 t[13] = -t[13];
|
|
190 t[15] = -t[15];
|
|
191 t += 16;
|
|
192 } while (t != t1);
|
|
193
|
|
194
|
|
195 t = tab;
|
|
196 t1 = tab + 8;
|
|
197 do {
|
|
198 int x1, x2, x3, x4;
|
|
199
|
|
200 x3 = MUL(t[16], FIX(SQRT2*0.5));
|
|
201 x4 = t[0] - x3;
|
|
202 x3 = t[0] + x3;
|
|
203
|
|
204 x2 = MUL(-(t[24] + t[8]), FIX(SQRT2*0.5));
|
|
205 x1 = MUL((t[8] - x2), xp[0]);
|
|
206 x2 = MUL((t[8] + x2), xp[1]);
|
|
207
|
|
208 t[ 0] = x3 + x1;
|
|
209 t[ 8] = x4 - x2;
|
|
210 t[16] = x4 + x2;
|
|
211 t[24] = x3 - x1;
|
|
212 t++;
|
|
213 } while (t != t1);
|
|
214
|
|
215 xp += 2;
|
|
216 t = tab;
|
|
217 t1 = tab + 4;
|
|
218 do {
|
|
219 xr = MUL(t[28],xp[0]);
|
|
220 t[28] = (t[0] - xr);
|
|
221 t[0] = (t[0] + xr);
|
|
222
|
|
223 xr = MUL(t[4],xp[1]);
|
|
224 t[ 4] = (t[24] - xr);
|
|
225 t[24] = (t[24] + xr);
|
|
226
|
|
227 xr = MUL(t[20],xp[2]);
|
|
228 t[20] = (t[8] - xr);
|
|
229 t[ 8] = (t[8] + xr);
|
|
230
|
|
231 xr = MUL(t[12],xp[3]);
|
|
232 t[12] = (t[16] - xr);
|
|
233 t[16] = (t[16] + xr);
|
|
234 t++;
|
|
235 } while (t != t1);
|
|
236 xp += 4;
|
|
237
|
|
238 for (i = 0; i < 4; i++) {
|
|
239 xr = MUL(tab[30-i*4],xp[0]);
|
|
240 tab[30-i*4] = (tab[i*4] - xr);
|
|
241 tab[ i*4] = (tab[i*4] + xr);
|
|
242
|
|
243 xr = MUL(tab[ 2+i*4],xp[1]);
|
|
244 tab[ 2+i*4] = (tab[28-i*4] - xr);
|
|
245 tab[28-i*4] = (tab[28-i*4] + xr);
|
|
246
|
|
247 xr = MUL(tab[31-i*4],xp[0]);
|
|
248 tab[31-i*4] = (tab[1+i*4] - xr);
|
|
249 tab[ 1+i*4] = (tab[1+i*4] + xr);
|
|
250
|
|
251 xr = MUL(tab[ 3+i*4],xp[1]);
|
|
252 tab[ 3+i*4] = (tab[29-i*4] - xr);
|
|
253 tab[29-i*4] = (tab[29-i*4] + xr);
|
|
254
|
|
255 xp += 2;
|
|
256 }
|
|
257
|
|
258 t = tab + 30;
|
|
259 t1 = tab + 1;
|
|
260 do {
|
|
261 xr = MUL(t1[0], *xp);
|
|
262 t1[0] = (t[0] - xr);
|
|
263 t[0] = (t[0] + xr);
|
|
264 t -= 2;
|
|
265 t1 += 2;
|
|
266 xp++;
|
|
267 } while (t >= tab);
|
|
268
|
|
269 for(i=0;i<32;i++) {
|
|
270 out[i] = tab[bitinv32[i]] << left_shift;
|
|
271 }
|
|
272 }
|
|
273
|
|
274 static void filter(MpegAudioContext *s, int ch, short *samples, int incr)
|
|
275 {
|
|
276 short *p, *q;
|
|
277 int sum, offset, i, j, norm, n;
|
|
278 short tmp[64];
|
|
279 int tmp1[32];
|
|
280 int *out;
|
|
281
|
|
282 // print_pow1(samples, 1152);
|
|
283
|
|
284 offset = s->samples_offset[ch];
|
|
285 out = &s->sb_samples[ch][0][0][0];
|
|
286 for(j=0;j<36;j++) {
|
|
287 /* 32 samples at once */
|
|
288 for(i=0;i<32;i++) {
|
|
289 s->samples_buf[ch][offset + (31 - i)] = samples[0];
|
|
290 samples += incr;
|
|
291 }
|
|
292
|
|
293 /* filter */
|
|
294 p = s->samples_buf[ch] + offset;
|
|
295 q = filter_bank;
|
|
296 /* maxsum = 23169 */
|
|
297 for(i=0;i<64;i++) {
|
|
298 sum = p[0*64] * q[0*64];
|
|
299 sum += p[1*64] * q[1*64];
|
|
300 sum += p[2*64] * q[2*64];
|
|
301 sum += p[3*64] * q[3*64];
|
|
302 sum += p[4*64] * q[4*64];
|
|
303 sum += p[5*64] * q[5*64];
|
|
304 sum += p[6*64] * q[6*64];
|
|
305 sum += p[7*64] * q[7*64];
|
|
306 tmp[i] = sum >> 14;
|
|
307 p++;
|
|
308 q++;
|
|
309 }
|
|
310 tmp1[0] = tmp[16];
|
|
311 for( i=1; i<=16; i++ ) tmp1[i] = tmp[i+16]+tmp[16-i];
|
|
312 for( i=17; i<=31; i++ ) tmp1[i] = tmp[i+16]-tmp[80-i];
|
|
313
|
|
314 /* integer IDCT 32 with normalization. XXX: There may be some
|
|
315 overflow left */
|
|
316 norm = 0;
|
|
317 for(i=0;i<32;i++) {
|
|
318 norm |= abs(tmp1[i]);
|
|
319 }
|
70
|
320 n = av_log2(norm) - 12;
|
0
|
321 if (n > 0) {
|
|
322 for(i=0;i<32;i++)
|
|
323 tmp1[i] >>= n;
|
|
324 } else {
|
|
325 n = 0;
|
|
326 }
|
|
327
|
|
328 idct32(out, tmp1, s->sblimit, n);
|
|
329
|
|
330 /* advance of 32 samples */
|
|
331 offset -= 32;
|
|
332 out += 32;
|
|
333 /* handle the wrap around */
|
|
334 if (offset < 0) {
|
|
335 memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32),
|
|
336 s->samples_buf[ch], (512 - 32) * 2);
|
|
337 offset = SAMPLES_BUF_SIZE - 512;
|
|
338 }
|
|
339 }
|
|
340 s->samples_offset[ch] = offset;
|
|
341
|
|
342 // print_pow(s->sb_samples, 1152);
|
|
343 }
|
|
344
|
|
345 static void compute_scale_factors(unsigned char scale_code[SBLIMIT],
|
|
346 unsigned char scale_factors[SBLIMIT][3],
|
|
347 int sb_samples[3][12][SBLIMIT],
|
|
348 int sblimit)
|
|
349 {
|
|
350 int *p, vmax, v, n, i, j, k, code;
|
|
351 int index, d1, d2;
|
|
352 unsigned char *sf = &scale_factors[0][0];
|
|
353
|
|
354 for(j=0;j<sblimit;j++) {
|
|
355 for(i=0;i<3;i++) {
|
|
356 /* find the max absolute value */
|
|
357 p = &sb_samples[i][0][j];
|
|
358 vmax = abs(*p);
|
|
359 for(k=1;k<12;k++) {
|
|
360 p += SBLIMIT;
|
|
361 v = abs(*p);
|
|
362 if (v > vmax)
|
|
363 vmax = v;
|
|
364 }
|
|
365 /* compute the scale factor index using log 2 computations */
|
|
366 if (vmax > 0) {
|
70
|
367 n = av_log2(vmax);
|
0
|
368 /* n is the position of the MSB of vmax. now
|
|
369 use at most 2 compares to find the index */
|
|
370 index = (21 - n) * 3 - 3;
|
|
371 if (index >= 0) {
|
|
372 while (vmax <= scale_factor_table[index+1])
|
|
373 index++;
|
|
374 } else {
|
|
375 index = 0; /* very unlikely case of overflow */
|
|
376 }
|
|
377 } else {
|
|
378 index = 63;
|
|
379 }
|
|
380
|
|
381 #if 0
|
|
382 printf("%2d:%d in=%x %x %d\n",
|
|
383 j, i, vmax, scale_factor_table[index], index);
|
|
384 #endif
|
|
385 /* store the scale factor */
|
|
386 assert(index >=0 && index <= 63);
|
|
387 sf[i] = index;
|
|
388 }
|
|
389
|
|
390 /* compute the transmission factor : look if the scale factors
|
|
391 are close enough to each other */
|
|
392 d1 = scale_diff_table[sf[0] - sf[1] + 64];
|
|
393 d2 = scale_diff_table[sf[1] - sf[2] + 64];
|
|
394
|
|
395 /* handle the 25 cases */
|
|
396 switch(d1 * 5 + d2) {
|
|
397 case 0*5+0:
|
|
398 case 0*5+4:
|
|
399 case 3*5+4:
|
|
400 case 4*5+0:
|
|
401 case 4*5+4:
|
|
402 code = 0;
|
|
403 break;
|
|
404 case 0*5+1:
|
|
405 case 0*5+2:
|
|
406 case 4*5+1:
|
|
407 case 4*5+2:
|
|
408 code = 3;
|
|
409 sf[2] = sf[1];
|
|
410 break;
|
|
411 case 0*5+3:
|
|
412 case 4*5+3:
|
|
413 code = 3;
|
|
414 sf[1] = sf[2];
|
|
415 break;
|
|
416 case 1*5+0:
|
|
417 case 1*5+4:
|
|
418 case 2*5+4:
|
|
419 code = 1;
|
|
420 sf[1] = sf[0];
|
|
421 break;
|
|
422 case 1*5+1:
|
|
423 case 1*5+2:
|
|
424 case 2*5+0:
|
|
425 case 2*5+1:
|
|
426 case 2*5+2:
|
|
427 code = 2;
|
|
428 sf[1] = sf[2] = sf[0];
|
|
429 break;
|
|
430 case 2*5+3:
|
|
431 case 3*5+3:
|
|
432 code = 2;
|
|
433 sf[0] = sf[1] = sf[2];
|
|
434 break;
|
|
435 case 3*5+0:
|
|
436 case 3*5+1:
|
|
437 case 3*5+2:
|
|
438 code = 2;
|
|
439 sf[0] = sf[2] = sf[1];
|
|
440 break;
|
|
441 case 1*5+3:
|
|
442 code = 2;
|
|
443 if (sf[0] > sf[2])
|
|
444 sf[0] = sf[2];
|
|
445 sf[1] = sf[2] = sf[0];
|
|
446 break;
|
|
447 default:
|
|
448 abort();
|
|
449 }
|
|
450
|
|
451 #if 0
|
|
452 printf("%d: %2d %2d %2d %d %d -> %d\n", j,
|
|
453 sf[0], sf[1], sf[2], d1, d2, code);
|
|
454 #endif
|
|
455 scale_code[j] = code;
|
|
456 sf += 3;
|
|
457 }
|
|
458 }
|
|
459
|
|
460 /* The most important function : psycho acoustic module. In this
|
|
461 encoder there is basically none, so this is the worst you can do,
|
|
462 but also this is the simpler. */
|
|
463 static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT])
|
|
464 {
|
|
465 int i;
|
|
466
|
|
467 for(i=0;i<s->sblimit;i++) {
|
|
468 smr[i] = (int)(fixed_smr[i] * 10);
|
|
469 }
|
|
470 }
|
|
471
|
|
472
|
|
473 #define SB_NOTALLOCATED 0
|
|
474 #define SB_ALLOCATED 1
|
|
475 #define SB_NOMORE 2
|
|
476
|
|
477 /* Try to maximize the smr while using a number of bits inferior to
|
|
478 the frame size. I tried to make the code simpler, faster and
|
|
479 smaller than other encoders :-) */
|
|
480 static void compute_bit_allocation(MpegAudioContext *s,
|
|
481 short smr1[MPA_MAX_CHANNELS][SBLIMIT],
|
|
482 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
|
|
483 int *padding)
|
|
484 {
|
|
485 int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size;
|
|
486 int incr;
|
|
487 short smr[MPA_MAX_CHANNELS][SBLIMIT];
|
|
488 unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT];
|
|
489 const unsigned char *alloc;
|
|
490
|
|
491 memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT);
|
|
492 memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT);
|
|
493 memset(bit_alloc, 0, s->nb_channels * SBLIMIT);
|
|
494
|
|
495 /* compute frame size and padding */
|
|
496 max_frame_size = s->frame_size;
|
|
497 s->frame_frac += s->frame_frac_incr;
|
|
498 if (s->frame_frac >= 65536) {
|
|
499 s->frame_frac -= 65536;
|
|
500 s->do_padding = 1;
|
|
501 max_frame_size += 8;
|
|
502 } else {
|
|
503 s->do_padding = 0;
|
|
504 }
|
|
505
|
|
506 /* compute the header + bit alloc size */
|
|
507 current_frame_size = 32;
|
|
508 alloc = s->alloc_table;
|
|
509 for(i=0;i<s->sblimit;i++) {
|
|
510 incr = alloc[0];
|
|
511 current_frame_size += incr * s->nb_channels;
|
|
512 alloc += 1 << incr;
|
|
513 }
|
|
514 for(;;) {
|
|
515 /* look for the subband with the largest signal to mask ratio */
|
|
516 max_sb = -1;
|
|
517 max_ch = -1;
|
|
518 max_smr = 0x80000000;
|
|
519 for(ch=0;ch<s->nb_channels;ch++) {
|
|
520 for(i=0;i<s->sblimit;i++) {
|
|
521 if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) {
|
|
522 max_smr = smr[ch][i];
|
|
523 max_sb = i;
|
|
524 max_ch = ch;
|
|
525 }
|
|
526 }
|
|
527 }
|
|
528 #if 0
|
|
529 printf("current=%d max=%d max_sb=%d alloc=%d\n",
|
|
530 current_frame_size, max_frame_size, max_sb,
|
|
531 bit_alloc[max_sb]);
|
|
532 #endif
|
|
533 if (max_sb < 0)
|
|
534 break;
|
|
535
|
|
536 /* find alloc table entry (XXX: not optimal, should use
|
|
537 pointer table) */
|
|
538 alloc = s->alloc_table;
|
|
539 for(i=0;i<max_sb;i++) {
|
|
540 alloc += 1 << alloc[0];
|
|
541 }
|
|
542
|
|
543 if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) {
|
|
544 /* nothing was coded for this band: add the necessary bits */
|
|
545 incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6;
|
|
546 incr += total_quant_bits[alloc[1]];
|
|
547 } else {
|
|
548 /* increments bit allocation */
|
|
549 b = bit_alloc[max_ch][max_sb];
|
|
550 incr = total_quant_bits[alloc[b + 1]] -
|
|
551 total_quant_bits[alloc[b]];
|
|
552 }
|
|
553
|
|
554 if (current_frame_size + incr <= max_frame_size) {
|
|
555 /* can increase size */
|
|
556 b = ++bit_alloc[max_ch][max_sb];
|
|
557 current_frame_size += incr;
|
|
558 /* decrease smr by the resolution we added */
|
|
559 smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]];
|
|
560 /* max allocation size reached ? */
|
|
561 if (b == ((1 << alloc[0]) - 1))
|
|
562 subband_status[max_ch][max_sb] = SB_NOMORE;
|
|
563 else
|
|
564 subband_status[max_ch][max_sb] = SB_ALLOCATED;
|
|
565 } else {
|
|
566 /* cannot increase the size of this subband */
|
|
567 subband_status[max_ch][max_sb] = SB_NOMORE;
|
|
568 }
|
|
569 }
|
|
570 *padding = max_frame_size - current_frame_size;
|
|
571 assert(*padding >= 0);
|
|
572
|
|
573 #if 0
|
|
574 for(i=0;i<s->sblimit;i++) {
|
|
575 printf("%d ", bit_alloc[i]);
|
|
576 }
|
|
577 printf("\n");
|
|
578 #endif
|
|
579 }
|
|
580
|
|
581 /*
|
|
582 * Output the mpeg audio layer 2 frame. Note how the code is small
|
|
583 * compared to other encoders :-)
|
|
584 */
|
|
585 static void encode_frame(MpegAudioContext *s,
|
|
586 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
|
|
587 int padding)
|
|
588 {
|
|
589 int i, j, k, l, bit_alloc_bits, b, ch;
|
|
590 unsigned char *sf;
|
|
591 int q[3];
|
|
592 PutBitContext *p = &s->pb;
|
|
593
|
|
594 /* header */
|
|
595
|
|
596 put_bits(p, 12, 0xfff);
|
|
597 put_bits(p, 1, 1 - s->lsf); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */
|
|
598 put_bits(p, 2, 4-2); /* layer 2 */
|
|
599 put_bits(p, 1, 1); /* no error protection */
|
|
600 put_bits(p, 4, s->bitrate_index);
|
|
601 put_bits(p, 2, s->freq_index);
|
|
602 put_bits(p, 1, s->do_padding); /* use padding */
|
|
603 put_bits(p, 1, 0); /* private_bit */
|
|
604 put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO);
|
|
605 put_bits(p, 2, 0); /* mode_ext */
|
|
606 put_bits(p, 1, 0); /* no copyright */
|
|
607 put_bits(p, 1, 1); /* original */
|
|
608 put_bits(p, 2, 0); /* no emphasis */
|
|
609
|
|
610 /* bit allocation */
|
|
611 j = 0;
|
|
612 for(i=0;i<s->sblimit;i++) {
|
|
613 bit_alloc_bits = s->alloc_table[j];
|
|
614 for(ch=0;ch<s->nb_channels;ch++) {
|
|
615 put_bits(p, bit_alloc_bits, bit_alloc[ch][i]);
|
|
616 }
|
|
617 j += 1 << bit_alloc_bits;
|
|
618 }
|
|
619
|
|
620 /* scale codes */
|
|
621 for(i=0;i<s->sblimit;i++) {
|
|
622 for(ch=0;ch<s->nb_channels;ch++) {
|
|
623 if (bit_alloc[ch][i])
|
|
624 put_bits(p, 2, s->scale_code[ch][i]);
|
|
625 }
|
|
626 }
|
|
627
|
|
628 /* scale factors */
|
|
629 for(i=0;i<s->sblimit;i++) {
|
|
630 for(ch=0;ch<s->nb_channels;ch++) {
|
|
631 if (bit_alloc[ch][i]) {
|
|
632 sf = &s->scale_factors[ch][i][0];
|
|
633 switch(s->scale_code[ch][i]) {
|
|
634 case 0:
|
|
635 put_bits(p, 6, sf[0]);
|
|
636 put_bits(p, 6, sf[1]);
|
|
637 put_bits(p, 6, sf[2]);
|
|
638 break;
|
|
639 case 3:
|
|
640 case 1:
|
|
641 put_bits(p, 6, sf[0]);
|
|
642 put_bits(p, 6, sf[2]);
|
|
643 break;
|
|
644 case 2:
|
|
645 put_bits(p, 6, sf[0]);
|
|
646 break;
|
|
647 }
|
|
648 }
|
|
649 }
|
|
650 }
|
|
651
|
|
652 /* quantization & write sub band samples */
|
|
653
|
|
654 for(k=0;k<3;k++) {
|
|
655 for(l=0;l<12;l+=3) {
|
|
656 j = 0;
|
|
657 for(i=0;i<s->sblimit;i++) {
|
|
658 bit_alloc_bits = s->alloc_table[j];
|
|
659 for(ch=0;ch<s->nb_channels;ch++) {
|
|
660 b = bit_alloc[ch][i];
|
|
661 if (b) {
|
|
662 int qindex, steps, m, sample, bits;
|
|
663 /* we encode 3 sub band samples of the same sub band at a time */
|
|
664 qindex = s->alloc_table[j+b];
|
|
665 steps = quant_steps[qindex];
|
|
666 for(m=0;m<3;m++) {
|
|
667 sample = s->sb_samples[ch][k][l + m][i];
|
|
668 /* divide by scale factor */
|
|
669 #ifdef USE_FLOATS
|
|
670 {
|
|
671 float a;
|
|
672 a = (float)sample * scale_factor_inv_table[s->scale_factors[ch][i][k]];
|
|
673 q[m] = (int)((a + 1.0) * steps * 0.5);
|
|
674 }
|
|
675 #else
|
|
676 {
|
|
677 int q1, e, shift, mult;
|
|
678 e = s->scale_factors[ch][i][k];
|
|
679 shift = scale_factor_shift[e];
|
|
680 mult = scale_factor_mult[e];
|
|
681
|
|
682 /* normalize to P bits */
|
|
683 if (shift < 0)
|
|
684 q1 = sample << (-shift);
|
|
685 else
|
|
686 q1 = sample >> shift;
|
|
687 q1 = (q1 * mult) >> P;
|
|
688 q[m] = ((q1 + (1 << P)) * steps) >> (P + 1);
|
|
689 }
|
|
690 #endif
|
|
691 if (q[m] >= steps)
|
|
692 q[m] = steps - 1;
|
|
693 assert(q[m] >= 0 && q[m] < steps);
|
|
694 }
|
|
695 bits = quant_bits[qindex];
|
|
696 if (bits < 0) {
|
|
697 /* group the 3 values to save bits */
|
|
698 put_bits(p, -bits,
|
|
699 q[0] + steps * (q[1] + steps * q[2]));
|
|
700 #if 0
|
|
701 printf("%d: gr1 %d\n",
|
|
702 i, q[0] + steps * (q[1] + steps * q[2]));
|
|
703 #endif
|
|
704 } else {
|
|
705 #if 0
|
|
706 printf("%d: gr3 %d %d %d\n",
|
|
707 i, q[0], q[1], q[2]);
|
|
708 #endif
|
|
709 put_bits(p, bits, q[0]);
|
|
710 put_bits(p, bits, q[1]);
|
|
711 put_bits(p, bits, q[2]);
|
|
712 }
|
|
713 }
|
|
714 }
|
|
715 /* next subband in alloc table */
|
|
716 j += 1 << bit_alloc_bits;
|
|
717 }
|
|
718 }
|
|
719 }
|
|
720
|
|
721 /* padding */
|
|
722 for(i=0;i<padding;i++)
|
|
723 put_bits(p, 1, 0);
|
|
724
|
|
725 /* flush */
|
|
726 flush_put_bits(p);
|
|
727 }
|
|
728
|
|
729 int MPA_encode_frame(AVCodecContext *avctx,
|
|
730 unsigned char *frame, int buf_size, void *data)
|
|
731 {
|
|
732 MpegAudioContext *s = avctx->priv_data;
|
|
733 short *samples = data;
|
|
734 short smr[MPA_MAX_CHANNELS][SBLIMIT];
|
|
735 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
|
|
736 int padding, i;
|
|
737
|
|
738 for(i=0;i<s->nb_channels;i++) {
|
|
739 filter(s, i, samples + i, s->nb_channels);
|
|
740 }
|
|
741
|
|
742 for(i=0;i<s->nb_channels;i++) {
|
|
743 compute_scale_factors(s->scale_code[i], s->scale_factors[i],
|
|
744 s->sb_samples[i], s->sblimit);
|
|
745 }
|
|
746 for(i=0;i<s->nb_channels;i++) {
|
|
747 psycho_acoustic_model(s, smr[i]);
|
|
748 }
|
|
749 compute_bit_allocation(s, smr, bit_alloc, &padding);
|
|
750
|
|
751 init_put_bits(&s->pb, frame, MPA_MAX_CODED_FRAME_SIZE, NULL, NULL);
|
|
752
|
|
753 encode_frame(s, bit_alloc, padding);
|
|
754
|
|
755 s->nb_samples += MPA_FRAME_SIZE;
|
|
756 return s->pb.buf_ptr - s->pb.buf;
|
|
757 }
|
|
758
|
|
759
|
|
760 AVCodec mp2_encoder = {
|
|
761 "mp2",
|
|
762 CODEC_TYPE_AUDIO,
|
|
763 CODEC_ID_MP2,
|
|
764 sizeof(MpegAudioContext),
|
|
765 MPA_encode_init,
|
|
766 MPA_encode_frame,
|
|
767 NULL,
|
|
768 };
|