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 */
|
|
19 #include <stdlib.h>
|
|
20 #include <stdio.h>
|
|
21 #include <string.h>
|
|
22 #include <math.h>
|
|
23 #include "avcodec.h"
|
|
24 #include "mpegaudio.h"
|
|
25
|
|
26 #define NDEBUG
|
|
27 #include <assert.h>
|
|
28
|
|
29 /* define it to use floats in quantization (I don't like floats !) */
|
|
30 //#define USE_FLOATS
|
|
31
|
|
32 #define MPA_STEREO 0
|
|
33 #define MPA_JSTEREO 1
|
|
34 #define MPA_DUAL 2
|
|
35 #define MPA_MONO 3
|
|
36
|
|
37 #include "mpegaudiotab.h"
|
|
38
|
|
39 int MPA_encode_init(AVCodecContext *avctx)
|
|
40 {
|
|
41 MpegAudioContext *s = avctx->priv_data;
|
|
42 int freq = avctx->sample_rate;
|
|
43 int bitrate = avctx->bit_rate;
|
|
44 int channels = avctx->channels;
|
|
45 int i, v, table, ch_bitrate;
|
|
46 float a;
|
|
47
|
|
48 if (channels > 2)
|
|
49 return -1;
|
|
50 bitrate = bitrate / 1000;
|
|
51 s->nb_channels = channels;
|
|
52 s->freq = freq;
|
|
53 s->bit_rate = bitrate * 1000;
|
|
54 avctx->frame_size = MPA_FRAME_SIZE;
|
|
55 avctx->key_frame = 1; /* always key frame */
|
|
56
|
|
57 /* encoding freq */
|
|
58 s->lsf = 0;
|
|
59 for(i=0;i<3;i++) {
|
|
60 if (freq_tab[i] == freq)
|
|
61 break;
|
|
62 if ((freq_tab[i] / 2) == freq) {
|
|
63 s->lsf = 1;
|
|
64 break;
|
|
65 }
|
|
66 }
|
|
67 if (i == 3)
|
|
68 return -1;
|
|
69 s->freq_index = i;
|
|
70
|
|
71 /* encoding bitrate & frequency */
|
|
72 for(i=0;i<15;i++) {
|
|
73 if (bitrate_tab[1-s->lsf][i] == bitrate)
|
|
74 break;
|
|
75 }
|
|
76 if (i == 15)
|
|
77 return -1;
|
|
78 s->bitrate_index = i;
|
|
79
|
|
80 /* compute total header size & pad bit */
|
|
81
|
|
82 a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0);
|
|
83 s->frame_size = ((int)a) * 8;
|
|
84
|
|
85 /* frame fractional size to compute padding */
|
|
86 s->frame_frac = 0;
|
|
87 s->frame_frac_incr = (int)((a - floor(a)) * 65536.0);
|
|
88
|
|
89 /* select the right allocation table */
|
|
90 ch_bitrate = bitrate / s->nb_channels;
|
|
91 if (!s->lsf) {
|
|
92 if ((freq == 48000 && ch_bitrate >= 56) ||
|
|
93 (ch_bitrate >= 56 && ch_bitrate <= 80))
|
|
94 table = 0;
|
|
95 else if (freq != 48000 && ch_bitrate >= 96)
|
|
96 table = 1;
|
|
97 else if (freq != 32000 && ch_bitrate <= 48)
|
|
98 table = 2;
|
|
99 else
|
|
100 table = 3;
|
|
101 } else {
|
|
102 table = 4;
|
|
103 }
|
|
104 /* number of used subbands */
|
|
105 s->sblimit = sblimit_table[table];
|
|
106 s->alloc_table = alloc_tables[table];
|
|
107
|
|
108 #ifdef DEBUG
|
|
109 printf("%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n",
|
|
110 bitrate, freq, s->frame_size, table, s->frame_frac_incr);
|
|
111 #endif
|
|
112
|
|
113 for(i=0;i<s->nb_channels;i++)
|
|
114 s->samples_offset[i] = 0;
|
|
115
|
|
116 for(i=0;i<512;i++) {
|
|
117 float a = enwindow[i] * 32768.0 * 16.0;
|
|
118 filter_bank[i] = (int)(a);
|
|
119 }
|
|
120 for(i=0;i<64;i++) {
|
|
121 v = (int)(pow(2.0, (3 - i) / 3.0) * (1 << 20));
|
|
122 if (v <= 0)
|
|
123 v = 1;
|
|
124 scale_factor_table[i] = v;
|
|
125 #ifdef USE_FLOATS
|
|
126 scale_factor_inv_table[i] = pow(2.0, -(3 - i) / 3.0) / (float)(1 << 20);
|
|
127 #else
|
|
128 #define P 15
|
|
129 scale_factor_shift[i] = 21 - P - (i / 3);
|
|
130 scale_factor_mult[i] = (1 << P) * pow(2.0, (i % 3) / 3.0);
|
|
131 #endif
|
|
132 }
|
|
133 for(i=0;i<128;i++) {
|
|
134 v = i - 64;
|
|
135 if (v <= -3)
|
|
136 v = 0;
|
|
137 else if (v < 0)
|
|
138 v = 1;
|
|
139 else if (v == 0)
|
|
140 v = 2;
|
|
141 else if (v < 3)
|
|
142 v = 3;
|
|
143 else
|
|
144 v = 4;
|
|
145 scale_diff_table[i] = v;
|
|
146 }
|
|
147
|
|
148 for(i=0;i<17;i++) {
|
|
149 v = quant_bits[i];
|
|
150 if (v < 0)
|
|
151 v = -v;
|
|
152 else
|
|
153 v = v * 3;
|
|
154 total_quant_bits[i] = 12 * v;
|
|
155 }
|
|
156
|
|
157 return 0;
|
|
158 }
|
|
159
|
|
160 /* 32 point floating point IDCT */
|
|
161 static void idct32(int *out, int *tab, int sblimit, int left_shift)
|
|
162 {
|
|
163 int i, j;
|
|
164 int *t, *t1, xr;
|
|
165 const int *xp = costab32;
|
|
166
|
|
167 for(j=31;j>=3;j-=2) tab[j] += tab[j - 2];
|
|
168
|
|
169 t = tab + 30;
|
|
170 t1 = tab + 2;
|
|
171 do {
|
|
172 t[0] += t[-4];
|
|
173 t[1] += t[1 - 4];
|
|
174 t -= 4;
|
|
175 } while (t != t1);
|
|
176
|
|
177 t = tab + 28;
|
|
178 t1 = tab + 4;
|
|
179 do {
|
|
180 t[0] += t[-8];
|
|
181 t[1] += t[1-8];
|
|
182 t[2] += t[2-8];
|
|
183 t[3] += t[3-8];
|
|
184 t -= 8;
|
|
185 } while (t != t1);
|
|
186
|
|
187 t = tab;
|
|
188 t1 = tab + 32;
|
|
189 do {
|
|
190 t[ 3] = -t[ 3];
|
|
191 t[ 6] = -t[ 6];
|
|
192
|
|
193 t[11] = -t[11];
|
|
194 t[12] = -t[12];
|
|
195 t[13] = -t[13];
|
|
196 t[15] = -t[15];
|
|
197 t += 16;
|
|
198 } while (t != t1);
|
|
199
|
|
200
|
|
201 t = tab;
|
|
202 t1 = tab + 8;
|
|
203 do {
|
|
204 int x1, x2, x3, x4;
|
|
205
|
|
206 x3 = MUL(t[16], FIX(SQRT2*0.5));
|
|
207 x4 = t[0] - x3;
|
|
208 x3 = t[0] + x3;
|
|
209
|
|
210 x2 = MUL(-(t[24] + t[8]), FIX(SQRT2*0.5));
|
|
211 x1 = MUL((t[8] - x2), xp[0]);
|
|
212 x2 = MUL((t[8] + x2), xp[1]);
|
|
213
|
|
214 t[ 0] = x3 + x1;
|
|
215 t[ 8] = x4 - x2;
|
|
216 t[16] = x4 + x2;
|
|
217 t[24] = x3 - x1;
|
|
218 t++;
|
|
219 } while (t != t1);
|
|
220
|
|
221 xp += 2;
|
|
222 t = tab;
|
|
223 t1 = tab + 4;
|
|
224 do {
|
|
225 xr = MUL(t[28],xp[0]);
|
|
226 t[28] = (t[0] - xr);
|
|
227 t[0] = (t[0] + xr);
|
|
228
|
|
229 xr = MUL(t[4],xp[1]);
|
|
230 t[ 4] = (t[24] - xr);
|
|
231 t[24] = (t[24] + xr);
|
|
232
|
|
233 xr = MUL(t[20],xp[2]);
|
|
234 t[20] = (t[8] - xr);
|
|
235 t[ 8] = (t[8] + xr);
|
|
236
|
|
237 xr = MUL(t[12],xp[3]);
|
|
238 t[12] = (t[16] - xr);
|
|
239 t[16] = (t[16] + xr);
|
|
240 t++;
|
|
241 } while (t != t1);
|
|
242 xp += 4;
|
|
243
|
|
244 for (i = 0; i < 4; i++) {
|
|
245 xr = MUL(tab[30-i*4],xp[0]);
|
|
246 tab[30-i*4] = (tab[i*4] - xr);
|
|
247 tab[ i*4] = (tab[i*4] + xr);
|
|
248
|
|
249 xr = MUL(tab[ 2+i*4],xp[1]);
|
|
250 tab[ 2+i*4] = (tab[28-i*4] - xr);
|
|
251 tab[28-i*4] = (tab[28-i*4] + xr);
|
|
252
|
|
253 xr = MUL(tab[31-i*4],xp[0]);
|
|
254 tab[31-i*4] = (tab[1+i*4] - xr);
|
|
255 tab[ 1+i*4] = (tab[1+i*4] + xr);
|
|
256
|
|
257 xr = MUL(tab[ 3+i*4],xp[1]);
|
|
258 tab[ 3+i*4] = (tab[29-i*4] - xr);
|
|
259 tab[29-i*4] = (tab[29-i*4] + xr);
|
|
260
|
|
261 xp += 2;
|
|
262 }
|
|
263
|
|
264 t = tab + 30;
|
|
265 t1 = tab + 1;
|
|
266 do {
|
|
267 xr = MUL(t1[0], *xp);
|
|
268 t1[0] = (t[0] - xr);
|
|
269 t[0] = (t[0] + xr);
|
|
270 t -= 2;
|
|
271 t1 += 2;
|
|
272 xp++;
|
|
273 } while (t >= tab);
|
|
274
|
|
275 for(i=0;i<32;i++) {
|
|
276 out[i] = tab[bitinv32[i]] << left_shift;
|
|
277 }
|
|
278 }
|
|
279
|
|
280 static void filter(MpegAudioContext *s, int ch, short *samples, int incr)
|
|
281 {
|
|
282 short *p, *q;
|
|
283 int sum, offset, i, j, norm, n;
|
|
284 short tmp[64];
|
|
285 int tmp1[32];
|
|
286 int *out;
|
|
287
|
|
288 // print_pow1(samples, 1152);
|
|
289
|
|
290 offset = s->samples_offset[ch];
|
|
291 out = &s->sb_samples[ch][0][0][0];
|
|
292 for(j=0;j<36;j++) {
|
|
293 /* 32 samples at once */
|
|
294 for(i=0;i<32;i++) {
|
|
295 s->samples_buf[ch][offset + (31 - i)] = samples[0];
|
|
296 samples += incr;
|
|
297 }
|
|
298
|
|
299 /* filter */
|
|
300 p = s->samples_buf[ch] + offset;
|
|
301 q = filter_bank;
|
|
302 /* maxsum = 23169 */
|
|
303 for(i=0;i<64;i++) {
|
|
304 sum = p[0*64] * q[0*64];
|
|
305 sum += p[1*64] * q[1*64];
|
|
306 sum += p[2*64] * q[2*64];
|
|
307 sum += p[3*64] * q[3*64];
|
|
308 sum += p[4*64] * q[4*64];
|
|
309 sum += p[5*64] * q[5*64];
|
|
310 sum += p[6*64] * q[6*64];
|
|
311 sum += p[7*64] * q[7*64];
|
|
312 tmp[i] = sum >> 14;
|
|
313 p++;
|
|
314 q++;
|
|
315 }
|
|
316 tmp1[0] = tmp[16];
|
|
317 for( i=1; i<=16; i++ ) tmp1[i] = tmp[i+16]+tmp[16-i];
|
|
318 for( i=17; i<=31; i++ ) tmp1[i] = tmp[i+16]-tmp[80-i];
|
|
319
|
|
320 /* integer IDCT 32 with normalization. XXX: There may be some
|
|
321 overflow left */
|
|
322 norm = 0;
|
|
323 for(i=0;i<32;i++) {
|
|
324 norm |= abs(tmp1[i]);
|
|
325 }
|
|
326 n = log2(norm) - 12;
|
|
327 if (n > 0) {
|
|
328 for(i=0;i<32;i++)
|
|
329 tmp1[i] >>= n;
|
|
330 } else {
|
|
331 n = 0;
|
|
332 }
|
|
333
|
|
334 idct32(out, tmp1, s->sblimit, n);
|
|
335
|
|
336 /* advance of 32 samples */
|
|
337 offset -= 32;
|
|
338 out += 32;
|
|
339 /* handle the wrap around */
|
|
340 if (offset < 0) {
|
|
341 memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32),
|
|
342 s->samples_buf[ch], (512 - 32) * 2);
|
|
343 offset = SAMPLES_BUF_SIZE - 512;
|
|
344 }
|
|
345 }
|
|
346 s->samples_offset[ch] = offset;
|
|
347
|
|
348 // print_pow(s->sb_samples, 1152);
|
|
349 }
|
|
350
|
|
351 static void compute_scale_factors(unsigned char scale_code[SBLIMIT],
|
|
352 unsigned char scale_factors[SBLIMIT][3],
|
|
353 int sb_samples[3][12][SBLIMIT],
|
|
354 int sblimit)
|
|
355 {
|
|
356 int *p, vmax, v, n, i, j, k, code;
|
|
357 int index, d1, d2;
|
|
358 unsigned char *sf = &scale_factors[0][0];
|
|
359
|
|
360 for(j=0;j<sblimit;j++) {
|
|
361 for(i=0;i<3;i++) {
|
|
362 /* find the max absolute value */
|
|
363 p = &sb_samples[i][0][j];
|
|
364 vmax = abs(*p);
|
|
365 for(k=1;k<12;k++) {
|
|
366 p += SBLIMIT;
|
|
367 v = abs(*p);
|
|
368 if (v > vmax)
|
|
369 vmax = v;
|
|
370 }
|
|
371 /* compute the scale factor index using log 2 computations */
|
|
372 if (vmax > 0) {
|
|
373 n = log2(vmax);
|
|
374 /* n is the position of the MSB of vmax. now
|
|
375 use at most 2 compares to find the index */
|
|
376 index = (21 - n) * 3 - 3;
|
|
377 if (index >= 0) {
|
|
378 while (vmax <= scale_factor_table[index+1])
|
|
379 index++;
|
|
380 } else {
|
|
381 index = 0; /* very unlikely case of overflow */
|
|
382 }
|
|
383 } else {
|
|
384 index = 63;
|
|
385 }
|
|
386
|
|
387 #if 0
|
|
388 printf("%2d:%d in=%x %x %d\n",
|
|
389 j, i, vmax, scale_factor_table[index], index);
|
|
390 #endif
|
|
391 /* store the scale factor */
|
|
392 assert(index >=0 && index <= 63);
|
|
393 sf[i] = index;
|
|
394 }
|
|
395
|
|
396 /* compute the transmission factor : look if the scale factors
|
|
397 are close enough to each other */
|
|
398 d1 = scale_diff_table[sf[0] - sf[1] + 64];
|
|
399 d2 = scale_diff_table[sf[1] - sf[2] + 64];
|
|
400
|
|
401 /* handle the 25 cases */
|
|
402 switch(d1 * 5 + d2) {
|
|
403 case 0*5+0:
|
|
404 case 0*5+4:
|
|
405 case 3*5+4:
|
|
406 case 4*5+0:
|
|
407 case 4*5+4:
|
|
408 code = 0;
|
|
409 break;
|
|
410 case 0*5+1:
|
|
411 case 0*5+2:
|
|
412 case 4*5+1:
|
|
413 case 4*5+2:
|
|
414 code = 3;
|
|
415 sf[2] = sf[1];
|
|
416 break;
|
|
417 case 0*5+3:
|
|
418 case 4*5+3:
|
|
419 code = 3;
|
|
420 sf[1] = sf[2];
|
|
421 break;
|
|
422 case 1*5+0:
|
|
423 case 1*5+4:
|
|
424 case 2*5+4:
|
|
425 code = 1;
|
|
426 sf[1] = sf[0];
|
|
427 break;
|
|
428 case 1*5+1:
|
|
429 case 1*5+2:
|
|
430 case 2*5+0:
|
|
431 case 2*5+1:
|
|
432 case 2*5+2:
|
|
433 code = 2;
|
|
434 sf[1] = sf[2] = sf[0];
|
|
435 break;
|
|
436 case 2*5+3:
|
|
437 case 3*5+3:
|
|
438 code = 2;
|
|
439 sf[0] = sf[1] = sf[2];
|
|
440 break;
|
|
441 case 3*5+0:
|
|
442 case 3*5+1:
|
|
443 case 3*5+2:
|
|
444 code = 2;
|
|
445 sf[0] = sf[2] = sf[1];
|
|
446 break;
|
|
447 case 1*5+3:
|
|
448 code = 2;
|
|
449 if (sf[0] > sf[2])
|
|
450 sf[0] = sf[2];
|
|
451 sf[1] = sf[2] = sf[0];
|
|
452 break;
|
|
453 default:
|
|
454 abort();
|
|
455 }
|
|
456
|
|
457 #if 0
|
|
458 printf("%d: %2d %2d %2d %d %d -> %d\n", j,
|
|
459 sf[0], sf[1], sf[2], d1, d2, code);
|
|
460 #endif
|
|
461 scale_code[j] = code;
|
|
462 sf += 3;
|
|
463 }
|
|
464 }
|
|
465
|
|
466 /* The most important function : psycho acoustic module. In this
|
|
467 encoder there is basically none, so this is the worst you can do,
|
|
468 but also this is the simpler. */
|
|
469 static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT])
|
|
470 {
|
|
471 int i;
|
|
472
|
|
473 for(i=0;i<s->sblimit;i++) {
|
|
474 smr[i] = (int)(fixed_smr[i] * 10);
|
|
475 }
|
|
476 }
|
|
477
|
|
478
|
|
479 #define SB_NOTALLOCATED 0
|
|
480 #define SB_ALLOCATED 1
|
|
481 #define SB_NOMORE 2
|
|
482
|
|
483 /* Try to maximize the smr while using a number of bits inferior to
|
|
484 the frame size. I tried to make the code simpler, faster and
|
|
485 smaller than other encoders :-) */
|
|
486 static void compute_bit_allocation(MpegAudioContext *s,
|
|
487 short smr1[MPA_MAX_CHANNELS][SBLIMIT],
|
|
488 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
|
|
489 int *padding)
|
|
490 {
|
|
491 int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size;
|
|
492 int incr;
|
|
493 short smr[MPA_MAX_CHANNELS][SBLIMIT];
|
|
494 unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT];
|
|
495 const unsigned char *alloc;
|
|
496
|
|
497 memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT);
|
|
498 memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT);
|
|
499 memset(bit_alloc, 0, s->nb_channels * SBLIMIT);
|
|
500
|
|
501 /* compute frame size and padding */
|
|
502 max_frame_size = s->frame_size;
|
|
503 s->frame_frac += s->frame_frac_incr;
|
|
504 if (s->frame_frac >= 65536) {
|
|
505 s->frame_frac -= 65536;
|
|
506 s->do_padding = 1;
|
|
507 max_frame_size += 8;
|
|
508 } else {
|
|
509 s->do_padding = 0;
|
|
510 }
|
|
511
|
|
512 /* compute the header + bit alloc size */
|
|
513 current_frame_size = 32;
|
|
514 alloc = s->alloc_table;
|
|
515 for(i=0;i<s->sblimit;i++) {
|
|
516 incr = alloc[0];
|
|
517 current_frame_size += incr * s->nb_channels;
|
|
518 alloc += 1 << incr;
|
|
519 }
|
|
520 for(;;) {
|
|
521 /* look for the subband with the largest signal to mask ratio */
|
|
522 max_sb = -1;
|
|
523 max_ch = -1;
|
|
524 max_smr = 0x80000000;
|
|
525 for(ch=0;ch<s->nb_channels;ch++) {
|
|
526 for(i=0;i<s->sblimit;i++) {
|
|
527 if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) {
|
|
528 max_smr = smr[ch][i];
|
|
529 max_sb = i;
|
|
530 max_ch = ch;
|
|
531 }
|
|
532 }
|
|
533 }
|
|
534 #if 0
|
|
535 printf("current=%d max=%d max_sb=%d alloc=%d\n",
|
|
536 current_frame_size, max_frame_size, max_sb,
|
|
537 bit_alloc[max_sb]);
|
|
538 #endif
|
|
539 if (max_sb < 0)
|
|
540 break;
|
|
541
|
|
542 /* find alloc table entry (XXX: not optimal, should use
|
|
543 pointer table) */
|
|
544 alloc = s->alloc_table;
|
|
545 for(i=0;i<max_sb;i++) {
|
|
546 alloc += 1 << alloc[0];
|
|
547 }
|
|
548
|
|
549 if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) {
|
|
550 /* nothing was coded for this band: add the necessary bits */
|
|
551 incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6;
|
|
552 incr += total_quant_bits[alloc[1]];
|
|
553 } else {
|
|
554 /* increments bit allocation */
|
|
555 b = bit_alloc[max_ch][max_sb];
|
|
556 incr = total_quant_bits[alloc[b + 1]] -
|
|
557 total_quant_bits[alloc[b]];
|
|
558 }
|
|
559
|
|
560 if (current_frame_size + incr <= max_frame_size) {
|
|
561 /* can increase size */
|
|
562 b = ++bit_alloc[max_ch][max_sb];
|
|
563 current_frame_size += incr;
|
|
564 /* decrease smr by the resolution we added */
|
|
565 smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]];
|
|
566 /* max allocation size reached ? */
|
|
567 if (b == ((1 << alloc[0]) - 1))
|
|
568 subband_status[max_ch][max_sb] = SB_NOMORE;
|
|
569 else
|
|
570 subband_status[max_ch][max_sb] = SB_ALLOCATED;
|
|
571 } else {
|
|
572 /* cannot increase the size of this subband */
|
|
573 subband_status[max_ch][max_sb] = SB_NOMORE;
|
|
574 }
|
|
575 }
|
|
576 *padding = max_frame_size - current_frame_size;
|
|
577 assert(*padding >= 0);
|
|
578
|
|
579 #if 0
|
|
580 for(i=0;i<s->sblimit;i++) {
|
|
581 printf("%d ", bit_alloc[i]);
|
|
582 }
|
|
583 printf("\n");
|
|
584 #endif
|
|
585 }
|
|
586
|
|
587 /*
|
|
588 * Output the mpeg audio layer 2 frame. Note how the code is small
|
|
589 * compared to other encoders :-)
|
|
590 */
|
|
591 static void encode_frame(MpegAudioContext *s,
|
|
592 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
|
|
593 int padding)
|
|
594 {
|
|
595 int i, j, k, l, bit_alloc_bits, b, ch;
|
|
596 unsigned char *sf;
|
|
597 int q[3];
|
|
598 PutBitContext *p = &s->pb;
|
|
599
|
|
600 /* header */
|
|
601
|
|
602 put_bits(p, 12, 0xfff);
|
|
603 put_bits(p, 1, 1 - s->lsf); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */
|
|
604 put_bits(p, 2, 4-2); /* layer 2 */
|
|
605 put_bits(p, 1, 1); /* no error protection */
|
|
606 put_bits(p, 4, s->bitrate_index);
|
|
607 put_bits(p, 2, s->freq_index);
|
|
608 put_bits(p, 1, s->do_padding); /* use padding */
|
|
609 put_bits(p, 1, 0); /* private_bit */
|
|
610 put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO);
|
|
611 put_bits(p, 2, 0); /* mode_ext */
|
|
612 put_bits(p, 1, 0); /* no copyright */
|
|
613 put_bits(p, 1, 1); /* original */
|
|
614 put_bits(p, 2, 0); /* no emphasis */
|
|
615
|
|
616 /* bit allocation */
|
|
617 j = 0;
|
|
618 for(i=0;i<s->sblimit;i++) {
|
|
619 bit_alloc_bits = s->alloc_table[j];
|
|
620 for(ch=0;ch<s->nb_channels;ch++) {
|
|
621 put_bits(p, bit_alloc_bits, bit_alloc[ch][i]);
|
|
622 }
|
|
623 j += 1 << bit_alloc_bits;
|
|
624 }
|
|
625
|
|
626 /* scale codes */
|
|
627 for(i=0;i<s->sblimit;i++) {
|
|
628 for(ch=0;ch<s->nb_channels;ch++) {
|
|
629 if (bit_alloc[ch][i])
|
|
630 put_bits(p, 2, s->scale_code[ch][i]);
|
|
631 }
|
|
632 }
|
|
633
|
|
634 /* scale factors */
|
|
635 for(i=0;i<s->sblimit;i++) {
|
|
636 for(ch=0;ch<s->nb_channels;ch++) {
|
|
637 if (bit_alloc[ch][i]) {
|
|
638 sf = &s->scale_factors[ch][i][0];
|
|
639 switch(s->scale_code[ch][i]) {
|
|
640 case 0:
|
|
641 put_bits(p, 6, sf[0]);
|
|
642 put_bits(p, 6, sf[1]);
|
|
643 put_bits(p, 6, sf[2]);
|
|
644 break;
|
|
645 case 3:
|
|
646 case 1:
|
|
647 put_bits(p, 6, sf[0]);
|
|
648 put_bits(p, 6, sf[2]);
|
|
649 break;
|
|
650 case 2:
|
|
651 put_bits(p, 6, sf[0]);
|
|
652 break;
|
|
653 }
|
|
654 }
|
|
655 }
|
|
656 }
|
|
657
|
|
658 /* quantization & write sub band samples */
|
|
659
|
|
660 for(k=0;k<3;k++) {
|
|
661 for(l=0;l<12;l+=3) {
|
|
662 j = 0;
|
|
663 for(i=0;i<s->sblimit;i++) {
|
|
664 bit_alloc_bits = s->alloc_table[j];
|
|
665 for(ch=0;ch<s->nb_channels;ch++) {
|
|
666 b = bit_alloc[ch][i];
|
|
667 if (b) {
|
|
668 int qindex, steps, m, sample, bits;
|
|
669 /* we encode 3 sub band samples of the same sub band at a time */
|
|
670 qindex = s->alloc_table[j+b];
|
|
671 steps = quant_steps[qindex];
|
|
672 for(m=0;m<3;m++) {
|
|
673 sample = s->sb_samples[ch][k][l + m][i];
|
|
674 /* divide by scale factor */
|
|
675 #ifdef USE_FLOATS
|
|
676 {
|
|
677 float a;
|
|
678 a = (float)sample * scale_factor_inv_table[s->scale_factors[ch][i][k]];
|
|
679 q[m] = (int)((a + 1.0) * steps * 0.5);
|
|
680 }
|
|
681 #else
|
|
682 {
|
|
683 int q1, e, shift, mult;
|
|
684 e = s->scale_factors[ch][i][k];
|
|
685 shift = scale_factor_shift[e];
|
|
686 mult = scale_factor_mult[e];
|
|
687
|
|
688 /* normalize to P bits */
|
|
689 if (shift < 0)
|
|
690 q1 = sample << (-shift);
|
|
691 else
|
|
692 q1 = sample >> shift;
|
|
693 q1 = (q1 * mult) >> P;
|
|
694 q[m] = ((q1 + (1 << P)) * steps) >> (P + 1);
|
|
695 }
|
|
696 #endif
|
|
697 if (q[m] >= steps)
|
|
698 q[m] = steps - 1;
|
|
699 assert(q[m] >= 0 && q[m] < steps);
|
|
700 }
|
|
701 bits = quant_bits[qindex];
|
|
702 if (bits < 0) {
|
|
703 /* group the 3 values to save bits */
|
|
704 put_bits(p, -bits,
|
|
705 q[0] + steps * (q[1] + steps * q[2]));
|
|
706 #if 0
|
|
707 printf("%d: gr1 %d\n",
|
|
708 i, q[0] + steps * (q[1] + steps * q[2]));
|
|
709 #endif
|
|
710 } else {
|
|
711 #if 0
|
|
712 printf("%d: gr3 %d %d %d\n",
|
|
713 i, q[0], q[1], q[2]);
|
|
714 #endif
|
|
715 put_bits(p, bits, q[0]);
|
|
716 put_bits(p, bits, q[1]);
|
|
717 put_bits(p, bits, q[2]);
|
|
718 }
|
|
719 }
|
|
720 }
|
|
721 /* next subband in alloc table */
|
|
722 j += 1 << bit_alloc_bits;
|
|
723 }
|
|
724 }
|
|
725 }
|
|
726
|
|
727 /* padding */
|
|
728 for(i=0;i<padding;i++)
|
|
729 put_bits(p, 1, 0);
|
|
730
|
|
731 /* flush */
|
|
732 flush_put_bits(p);
|
|
733 }
|
|
734
|
|
735 int MPA_encode_frame(AVCodecContext *avctx,
|
|
736 unsigned char *frame, int buf_size, void *data)
|
|
737 {
|
|
738 MpegAudioContext *s = avctx->priv_data;
|
|
739 short *samples = data;
|
|
740 short smr[MPA_MAX_CHANNELS][SBLIMIT];
|
|
741 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
|
|
742 int padding, i;
|
|
743
|
|
744 for(i=0;i<s->nb_channels;i++) {
|
|
745 filter(s, i, samples + i, s->nb_channels);
|
|
746 }
|
|
747
|
|
748 for(i=0;i<s->nb_channels;i++) {
|
|
749 compute_scale_factors(s->scale_code[i], s->scale_factors[i],
|
|
750 s->sb_samples[i], s->sblimit);
|
|
751 }
|
|
752 for(i=0;i<s->nb_channels;i++) {
|
|
753 psycho_acoustic_model(s, smr[i]);
|
|
754 }
|
|
755 compute_bit_allocation(s, smr, bit_alloc, &padding);
|
|
756
|
|
757 init_put_bits(&s->pb, frame, MPA_MAX_CODED_FRAME_SIZE, NULL, NULL);
|
|
758
|
|
759 encode_frame(s, bit_alloc, padding);
|
|
760
|
|
761 s->nb_samples += MPA_FRAME_SIZE;
|
|
762 return s->pb.buf_ptr - s->pb.buf;
|
|
763 }
|
|
764
|
|
765
|
|
766 AVCodec mp2_encoder = {
|
|
767 "mp2",
|
|
768 CODEC_TYPE_AUDIO,
|
|
769 CODEC_ID_MP2,
|
|
770 sizeof(MpegAudioContext),
|
|
771 MPA_encode_init,
|
|
772 MPA_encode_frame,
|
|
773 NULL,
|
|
774 };
|