Commit | Line | Data |
---|---|---|
de6d9b64 FB |
1 | /* |
2 | * The simplest mpeg audio layer 2 encoder | |
ff4ec49e | 3 | * Copyright (c) 2000, 2001 Fabrice Bellard. |
de6d9b64 | 4 | * |
ff4ec49e FB |
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. | |
de6d9b64 | 9 | * |
ff4ec49e | 10 | * This library is distributed in the hope that it will be useful, |
de6d9b64 | 11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
ff4ec49e FB |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
13 | * Lesser General Public License for more details. | |
de6d9b64 | 14 | * |
ff4ec49e FB |
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 | |
17 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
de6d9b64 | 18 | */ |
de6d9b64 FB |
19 | #include "avcodec.h" |
20 | #include "mpegaudio.h" | |
21 | ||
afa982fd FB |
22 | /* currently, cannot change these constants (need to modify |
23 | quantization stage) */ | |
24 | #define FRAC_BITS 15 | |
25 | #define WFRAC_BITS 14 | |
26 | #define MUL(a,b) (((INT64)(a) * (INT64)(b)) >> FRAC_BITS) | |
27 | #define FIX(a) ((int)((a) * (1 << FRAC_BITS))) | |
2456e28d FB |
28 | |
29 | #define SAMPLES_BUF_SIZE 4096 | |
30 | ||
31 | typedef struct MpegAudioContext { | |
32 | PutBitContext pb; | |
33 | int nb_channels; | |
34 | int freq, bit_rate; | |
35 | int lsf; /* 1 if mpeg2 low bitrate selected */ | |
36 | int bitrate_index; /* bit rate */ | |
37 | int freq_index; | |
38 | int frame_size; /* frame size, in bits, without padding */ | |
39 | INT64 nb_samples; /* total number of samples encoded */ | |
40 | /* padding computation */ | |
41 | int frame_frac, frame_frac_incr, do_padding; | |
42 | short samples_buf[MPA_MAX_CHANNELS][SAMPLES_BUF_SIZE]; /* buffer for filter */ | |
43 | int samples_offset[MPA_MAX_CHANNELS]; /* offset in samples_buf */ | |
44 | int sb_samples[MPA_MAX_CHANNELS][3][12][SBLIMIT]; | |
45 | unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3]; /* scale factors */ | |
46 | /* code to group 3 scale factors */ | |
47 | unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT]; | |
48 | int sblimit; /* number of used subbands */ | |
49 | const unsigned char *alloc_table; | |
50 | } MpegAudioContext; | |
51 | ||
de6d9b64 FB |
52 | /* define it to use floats in quantization (I don't like floats !) */ |
53 | //#define USE_FLOATS | |
54 | ||
de6d9b64 FB |
55 | #include "mpegaudiotab.h" |
56 | ||
57 | int MPA_encode_init(AVCodecContext *avctx) | |
58 | { | |
59 | MpegAudioContext *s = avctx->priv_data; | |
60 | int freq = avctx->sample_rate; | |
61 | int bitrate = avctx->bit_rate; | |
62 | int channels = avctx->channels; | |
2456e28d | 63 | int i, v, table; |
de6d9b64 FB |
64 | float a; |
65 | ||
66 | if (channels > 2) | |
67 | return -1; | |
68 | bitrate = bitrate / 1000; | |
69 | s->nb_channels = channels; | |
70 | s->freq = freq; | |
71 | s->bit_rate = bitrate * 1000; | |
72 | avctx->frame_size = MPA_FRAME_SIZE; | |
73 | avctx->key_frame = 1; /* always key frame */ | |
74 | ||
75 | /* encoding freq */ | |
76 | s->lsf = 0; | |
77 | for(i=0;i<3;i++) { | |
2456e28d | 78 | if (mpa_freq_tab[i] == freq) |
de6d9b64 | 79 | break; |
2456e28d | 80 | if ((mpa_freq_tab[i] / 2) == freq) { |
de6d9b64 FB |
81 | s->lsf = 1; |
82 | break; | |
83 | } | |
84 | } | |
85 | if (i == 3) | |
86 | return -1; | |
87 | s->freq_index = i; | |
88 | ||
89 | /* encoding bitrate & frequency */ | |
90 | for(i=0;i<15;i++) { | |
2456e28d | 91 | if (mpa_bitrate_tab[s->lsf][1][i] == bitrate) |
de6d9b64 FB |
92 | break; |
93 | } | |
94 | if (i == 15) | |
95 | return -1; | |
96 | s->bitrate_index = i; | |
97 | ||
98 | /* compute total header size & pad bit */ | |
99 | ||
100 | a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0); | |
101 | s->frame_size = ((int)a) * 8; | |
102 | ||
103 | /* frame fractional size to compute padding */ | |
104 | s->frame_frac = 0; | |
105 | s->frame_frac_incr = (int)((a - floor(a)) * 65536.0); | |
106 | ||
107 | /* select the right allocation table */ | |
2456e28d FB |
108 | table = l2_select_table(bitrate, s->nb_channels, freq, s->lsf); |
109 | ||
de6d9b64 FB |
110 | /* number of used subbands */ |
111 | s->sblimit = sblimit_table[table]; | |
112 | s->alloc_table = alloc_tables[table]; | |
113 | ||
114 | #ifdef DEBUG | |
115 | printf("%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n", | |
116 | bitrate, freq, s->frame_size, table, s->frame_frac_incr); | |
117 | #endif | |
118 | ||
119 | for(i=0;i<s->nb_channels;i++) | |
120 | s->samples_offset[i] = 0; | |
121 | ||
2456e28d FB |
122 | for(i=0;i<257;i++) { |
123 | int v; | |
afa982fd FB |
124 | v = mpa_enwindow[i]; |
125 | #if WFRAC_BITS != 16 | |
126 | v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS); | |
127 | #endif | |
2456e28d FB |
128 | filter_bank[i] = v; |
129 | if ((i & 63) != 0) | |
130 | v = -v; | |
131 | if (i != 0) | |
132 | filter_bank[512 - i] = v; | |
de6d9b64 | 133 | } |
2456e28d | 134 | |
de6d9b64 FB |
135 | for(i=0;i<64;i++) { |
136 | v = (int)(pow(2.0, (3 - i) / 3.0) * (1 << 20)); | |
137 | if (v <= 0) | |
138 | v = 1; | |
139 | scale_factor_table[i] = v; | |
140 | #ifdef USE_FLOATS | |
141 | scale_factor_inv_table[i] = pow(2.0, -(3 - i) / 3.0) / (float)(1 << 20); | |
142 | #else | |
143 | #define P 15 | |
144 | scale_factor_shift[i] = 21 - P - (i / 3); | |
145 | scale_factor_mult[i] = (1 << P) * pow(2.0, (i % 3) / 3.0); | |
146 | #endif | |
147 | } | |
148 | for(i=0;i<128;i++) { | |
149 | v = i - 64; | |
150 | if (v <= -3) | |
151 | v = 0; | |
152 | else if (v < 0) | |
153 | v = 1; | |
154 | else if (v == 0) | |
155 | v = 2; | |
156 | else if (v < 3) | |
157 | v = 3; | |
158 | else | |
159 | v = 4; | |
160 | scale_diff_table[i] = v; | |
161 | } | |
162 | ||
163 | for(i=0;i<17;i++) { | |
164 | v = quant_bits[i]; | |
165 | if (v < 0) | |
166 | v = -v; | |
167 | else | |
168 | v = v * 3; | |
169 | total_quant_bits[i] = 12 * v; | |
170 | } | |
171 | ||
172 | return 0; | |
173 | } | |
174 | ||
2456e28d | 175 | /* 32 point floating point IDCT without 1/sqrt(2) coef zero scaling */ |
afa982fd | 176 | static void idct32(int *out, int *tab) |
de6d9b64 FB |
177 | { |
178 | int i, j; | |
179 | int *t, *t1, xr; | |
180 | const int *xp = costab32; | |
181 | ||
182 | for(j=31;j>=3;j-=2) tab[j] += tab[j - 2]; | |
183 | ||
184 | t = tab + 30; | |
185 | t1 = tab + 2; | |
186 | do { | |
187 | t[0] += t[-4]; | |
188 | t[1] += t[1 - 4]; | |
189 | t -= 4; | |
190 | } while (t != t1); | |
191 | ||
192 | t = tab + 28; | |
193 | t1 = tab + 4; | |
194 | do { | |
195 | t[0] += t[-8]; | |
196 | t[1] += t[1-8]; | |
197 | t[2] += t[2-8]; | |
198 | t[3] += t[3-8]; | |
199 | t -= 8; | |
200 | } while (t != t1); | |
201 | ||
202 | t = tab; | |
203 | t1 = tab + 32; | |
204 | do { | |
205 | t[ 3] = -t[ 3]; | |
206 | t[ 6] = -t[ 6]; | |
207 | ||
208 | t[11] = -t[11]; | |
209 | t[12] = -t[12]; | |
210 | t[13] = -t[13]; | |
211 | t[15] = -t[15]; | |
212 | t += 16; | |
213 | } while (t != t1); | |
214 | ||
215 | ||
216 | t = tab; | |
217 | t1 = tab + 8; | |
218 | do { | |
219 | int x1, x2, x3, x4; | |
220 | ||
221 | x3 = MUL(t[16], FIX(SQRT2*0.5)); | |
222 | x4 = t[0] - x3; | |
223 | x3 = t[0] + x3; | |
224 | ||
225 | x2 = MUL(-(t[24] + t[8]), FIX(SQRT2*0.5)); | |
226 | x1 = MUL((t[8] - x2), xp[0]); | |
227 | x2 = MUL((t[8] + x2), xp[1]); | |
228 | ||
229 | t[ 0] = x3 + x1; | |
230 | t[ 8] = x4 - x2; | |
231 | t[16] = x4 + x2; | |
232 | t[24] = x3 - x1; | |
233 | t++; | |
234 | } while (t != t1); | |
235 | ||
236 | xp += 2; | |
237 | t = tab; | |
238 | t1 = tab + 4; | |
239 | do { | |
240 | xr = MUL(t[28],xp[0]); | |
241 | t[28] = (t[0] - xr); | |
242 | t[0] = (t[0] + xr); | |
243 | ||
244 | xr = MUL(t[4],xp[1]); | |
245 | t[ 4] = (t[24] - xr); | |
246 | t[24] = (t[24] + xr); | |
247 | ||
248 | xr = MUL(t[20],xp[2]); | |
249 | t[20] = (t[8] - xr); | |
250 | t[ 8] = (t[8] + xr); | |
251 | ||
252 | xr = MUL(t[12],xp[3]); | |
253 | t[12] = (t[16] - xr); | |
254 | t[16] = (t[16] + xr); | |
255 | t++; | |
256 | } while (t != t1); | |
257 | xp += 4; | |
258 | ||
259 | for (i = 0; i < 4; i++) { | |
260 | xr = MUL(tab[30-i*4],xp[0]); | |
261 | tab[30-i*4] = (tab[i*4] - xr); | |
262 | tab[ i*4] = (tab[i*4] + xr); | |
263 | ||
264 | xr = MUL(tab[ 2+i*4],xp[1]); | |
265 | tab[ 2+i*4] = (tab[28-i*4] - xr); | |
266 | tab[28-i*4] = (tab[28-i*4] + xr); | |
267 | ||
268 | xr = MUL(tab[31-i*4],xp[0]); | |
269 | tab[31-i*4] = (tab[1+i*4] - xr); | |
270 | tab[ 1+i*4] = (tab[1+i*4] + xr); | |
271 | ||
272 | xr = MUL(tab[ 3+i*4],xp[1]); | |
273 | tab[ 3+i*4] = (tab[29-i*4] - xr); | |
274 | tab[29-i*4] = (tab[29-i*4] + xr); | |
275 | ||
276 | xp += 2; | |
277 | } | |
278 | ||
279 | t = tab + 30; | |
280 | t1 = tab + 1; | |
281 | do { | |
282 | xr = MUL(t1[0], *xp); | |
283 | t1[0] = (t[0] - xr); | |
284 | t[0] = (t[0] + xr); | |
285 | t -= 2; | |
286 | t1 += 2; | |
287 | xp++; | |
288 | } while (t >= tab); | |
289 | ||
290 | for(i=0;i<32;i++) { | |
afa982fd | 291 | out[i] = tab[bitinv32[i]]; |
de6d9b64 FB |
292 | } |
293 | } | |
294 | ||
afa982fd FB |
295 | #define WSHIFT (WFRAC_BITS + 15 - FRAC_BITS) |
296 | ||
de6d9b64 FB |
297 | static void filter(MpegAudioContext *s, int ch, short *samples, int incr) |
298 | { | |
299 | short *p, *q; | |
afa982fd FB |
300 | int sum, offset, i, j; |
301 | int tmp[64]; | |
de6d9b64 FB |
302 | int tmp1[32]; |
303 | int *out; | |
304 | ||
305 | // print_pow1(samples, 1152); | |
306 | ||
307 | offset = s->samples_offset[ch]; | |
308 | out = &s->sb_samples[ch][0][0][0]; | |
309 | for(j=0;j<36;j++) { | |
310 | /* 32 samples at once */ | |
311 | for(i=0;i<32;i++) { | |
312 | s->samples_buf[ch][offset + (31 - i)] = samples[0]; | |
313 | samples += incr; | |
314 | } | |
315 | ||
316 | /* filter */ | |
317 | p = s->samples_buf[ch] + offset; | |
318 | q = filter_bank; | |
319 | /* maxsum = 23169 */ | |
320 | for(i=0;i<64;i++) { | |
321 | sum = p[0*64] * q[0*64]; | |
322 | sum += p[1*64] * q[1*64]; | |
323 | sum += p[2*64] * q[2*64]; | |
324 | sum += p[3*64] * q[3*64]; | |
325 | sum += p[4*64] * q[4*64]; | |
326 | sum += p[5*64] * q[5*64]; | |
327 | sum += p[6*64] * q[6*64]; | |
328 | sum += p[7*64] * q[7*64]; | |
afa982fd | 329 | tmp[i] = sum; |
de6d9b64 FB |
330 | p++; |
331 | q++; | |
332 | } | |
afa982fd FB |
333 | tmp1[0] = tmp[16] >> WSHIFT; |
334 | for( i=1; i<=16; i++ ) tmp1[i] = (tmp[i+16]+tmp[16-i]) >> WSHIFT; | |
335 | for( i=17; i<=31; i++ ) tmp1[i] = (tmp[i+16]-tmp[80-i]) >> WSHIFT; | |
de6d9b64 | 336 | |
afa982fd | 337 | idct32(out, tmp1); |
de6d9b64 FB |
338 | |
339 | /* advance of 32 samples */ | |
340 | offset -= 32; | |
341 | out += 32; | |
342 | /* handle the wrap around */ | |
343 | if (offset < 0) { | |
344 | memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32), | |
345 | s->samples_buf[ch], (512 - 32) * 2); | |
346 | offset = SAMPLES_BUF_SIZE - 512; | |
347 | } | |
348 | } | |
349 | s->samples_offset[ch] = offset; | |
350 | ||
351 | // print_pow(s->sb_samples, 1152); | |
352 | } | |
353 | ||
354 | static void compute_scale_factors(unsigned char scale_code[SBLIMIT], | |
355 | unsigned char scale_factors[SBLIMIT][3], | |
356 | int sb_samples[3][12][SBLIMIT], | |
357 | int sblimit) | |
358 | { | |
359 | int *p, vmax, v, n, i, j, k, code; | |
360 | int index, d1, d2; | |
361 | unsigned char *sf = &scale_factors[0][0]; | |
362 | ||
363 | for(j=0;j<sblimit;j++) { | |
364 | for(i=0;i<3;i++) { | |
365 | /* find the max absolute value */ | |
366 | p = &sb_samples[i][0][j]; | |
367 | vmax = abs(*p); | |
368 | for(k=1;k<12;k++) { | |
369 | p += SBLIMIT; | |
370 | v = abs(*p); | |
371 | if (v > vmax) | |
372 | vmax = v; | |
373 | } | |
374 | /* compute the scale factor index using log 2 computations */ | |
375 | if (vmax > 0) { | |
935442b5 | 376 | n = av_log2(vmax); |
de6d9b64 FB |
377 | /* n is the position of the MSB of vmax. now |
378 | use at most 2 compares to find the index */ | |
379 | index = (21 - n) * 3 - 3; | |
380 | if (index >= 0) { | |
381 | while (vmax <= scale_factor_table[index+1]) | |
382 | index++; | |
383 | } else { | |
384 | index = 0; /* very unlikely case of overflow */ | |
385 | } | |
386 | } else { | |
afa982fd | 387 | index = 62; /* value 63 is not allowed */ |
de6d9b64 | 388 | } |
afa982fd | 389 | |
de6d9b64 FB |
390 | #if 0 |
391 | printf("%2d:%d in=%x %x %d\n", | |
392 | j, i, vmax, scale_factor_table[index], index); | |
393 | #endif | |
394 | /* store the scale factor */ | |
395 | assert(index >=0 && index <= 63); | |
396 | sf[i] = index; | |
397 | } | |
398 | ||
399 | /* compute the transmission factor : look if the scale factors | |
400 | are close enough to each other */ | |
401 | d1 = scale_diff_table[sf[0] - sf[1] + 64]; | |
402 | d2 = scale_diff_table[sf[1] - sf[2] + 64]; | |
403 | ||
404 | /* handle the 25 cases */ | |
405 | switch(d1 * 5 + d2) { | |
406 | case 0*5+0: | |
407 | case 0*5+4: | |
408 | case 3*5+4: | |
409 | case 4*5+0: | |
410 | case 4*5+4: | |
411 | code = 0; | |
412 | break; | |
413 | case 0*5+1: | |
414 | case 0*5+2: | |
415 | case 4*5+1: | |
416 | case 4*5+2: | |
417 | code = 3; | |
418 | sf[2] = sf[1]; | |
419 | break; | |
420 | case 0*5+3: | |
421 | case 4*5+3: | |
422 | code = 3; | |
423 | sf[1] = sf[2]; | |
424 | break; | |
425 | case 1*5+0: | |
426 | case 1*5+4: | |
427 | case 2*5+4: | |
428 | code = 1; | |
429 | sf[1] = sf[0]; | |
430 | break; | |
431 | case 1*5+1: | |
432 | case 1*5+2: | |
433 | case 2*5+0: | |
434 | case 2*5+1: | |
435 | case 2*5+2: | |
436 | code = 2; | |
437 | sf[1] = sf[2] = sf[0]; | |
438 | break; | |
439 | case 2*5+3: | |
440 | case 3*5+3: | |
441 | code = 2; | |
442 | sf[0] = sf[1] = sf[2]; | |
443 | break; | |
444 | case 3*5+0: | |
445 | case 3*5+1: | |
446 | case 3*5+2: | |
447 | code = 2; | |
448 | sf[0] = sf[2] = sf[1]; | |
449 | break; | |
450 | case 1*5+3: | |
451 | code = 2; | |
452 | if (sf[0] > sf[2]) | |
453 | sf[0] = sf[2]; | |
454 | sf[1] = sf[2] = sf[0]; | |
455 | break; | |
456 | default: | |
02ac3136 | 457 | av_abort(); |
de6d9b64 FB |
458 | } |
459 | ||
460 | #if 0 | |
461 | printf("%d: %2d %2d %2d %d %d -> %d\n", j, | |
462 | sf[0], sf[1], sf[2], d1, d2, code); | |
463 | #endif | |
464 | scale_code[j] = code; | |
465 | sf += 3; | |
466 | } | |
467 | } | |
468 | ||
469 | /* The most important function : psycho acoustic module. In this | |
470 | encoder there is basically none, so this is the worst you can do, | |
471 | but also this is the simpler. */ | |
472 | static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT]) | |
473 | { | |
474 | int i; | |
475 | ||
476 | for(i=0;i<s->sblimit;i++) { | |
477 | smr[i] = (int)(fixed_smr[i] * 10); | |
478 | } | |
479 | } | |
480 | ||
481 | ||
482 | #define SB_NOTALLOCATED 0 | |
483 | #define SB_ALLOCATED 1 | |
484 | #define SB_NOMORE 2 | |
485 | ||
486 | /* Try to maximize the smr while using a number of bits inferior to | |
487 | the frame size. I tried to make the code simpler, faster and | |
488 | smaller than other encoders :-) */ | |
489 | static void compute_bit_allocation(MpegAudioContext *s, | |
490 | short smr1[MPA_MAX_CHANNELS][SBLIMIT], | |
491 | unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], | |
492 | int *padding) | |
493 | { | |
494 | int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size; | |
495 | int incr; | |
496 | short smr[MPA_MAX_CHANNELS][SBLIMIT]; | |
497 | unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT]; | |
498 | const unsigned char *alloc; | |
499 | ||
500 | memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT); | |
501 | memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT); | |
502 | memset(bit_alloc, 0, s->nb_channels * SBLIMIT); | |
503 | ||
504 | /* compute frame size and padding */ | |
505 | max_frame_size = s->frame_size; | |
506 | s->frame_frac += s->frame_frac_incr; | |
507 | if (s->frame_frac >= 65536) { | |
508 | s->frame_frac -= 65536; | |
509 | s->do_padding = 1; | |
510 | max_frame_size += 8; | |
511 | } else { | |
512 | s->do_padding = 0; | |
513 | } | |
514 | ||
515 | /* compute the header + bit alloc size */ | |
516 | current_frame_size = 32; | |
517 | alloc = s->alloc_table; | |
518 | for(i=0;i<s->sblimit;i++) { | |
519 | incr = alloc[0]; | |
520 | current_frame_size += incr * s->nb_channels; | |
521 | alloc += 1 << incr; | |
522 | } | |
523 | for(;;) { | |
524 | /* look for the subband with the largest signal to mask ratio */ | |
525 | max_sb = -1; | |
526 | max_ch = -1; | |
527 | max_smr = 0x80000000; | |
528 | for(ch=0;ch<s->nb_channels;ch++) { | |
529 | for(i=0;i<s->sblimit;i++) { | |
530 | if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) { | |
531 | max_smr = smr[ch][i]; | |
532 | max_sb = i; | |
533 | max_ch = ch; | |
534 | } | |
535 | } | |
536 | } | |
537 | #if 0 | |
538 | printf("current=%d max=%d max_sb=%d alloc=%d\n", | |
539 | current_frame_size, max_frame_size, max_sb, | |
540 | bit_alloc[max_sb]); | |
541 | #endif | |
542 | if (max_sb < 0) | |
543 | break; | |
544 | ||
545 | /* find alloc table entry (XXX: not optimal, should use | |
546 | pointer table) */ | |
547 | alloc = s->alloc_table; | |
548 | for(i=0;i<max_sb;i++) { | |
549 | alloc += 1 << alloc[0]; | |
550 | } | |
551 | ||
552 | if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) { | |
553 | /* nothing was coded for this band: add the necessary bits */ | |
554 | incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6; | |
555 | incr += total_quant_bits[alloc[1]]; | |
556 | } else { | |
557 | /* increments bit allocation */ | |
558 | b = bit_alloc[max_ch][max_sb]; | |
559 | incr = total_quant_bits[alloc[b + 1]] - | |
560 | total_quant_bits[alloc[b]]; | |
561 | } | |
562 | ||
563 | if (current_frame_size + incr <= max_frame_size) { | |
564 | /* can increase size */ | |
565 | b = ++bit_alloc[max_ch][max_sb]; | |
566 | current_frame_size += incr; | |
567 | /* decrease smr by the resolution we added */ | |
568 | smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]]; | |
569 | /* max allocation size reached ? */ | |
570 | if (b == ((1 << alloc[0]) - 1)) | |
571 | subband_status[max_ch][max_sb] = SB_NOMORE; | |
572 | else | |
573 | subband_status[max_ch][max_sb] = SB_ALLOCATED; | |
574 | } else { | |
575 | /* cannot increase the size of this subband */ | |
576 | subband_status[max_ch][max_sb] = SB_NOMORE; | |
577 | } | |
578 | } | |
579 | *padding = max_frame_size - current_frame_size; | |
580 | assert(*padding >= 0); | |
581 | ||
582 | #if 0 | |
583 | for(i=0;i<s->sblimit;i++) { | |
584 | printf("%d ", bit_alloc[i]); | |
585 | } | |
586 | printf("\n"); | |
587 | #endif | |
588 | } | |
589 | ||
590 | /* | |
591 | * Output the mpeg audio layer 2 frame. Note how the code is small | |
592 | * compared to other encoders :-) | |
593 | */ | |
594 | static void encode_frame(MpegAudioContext *s, | |
595 | unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], | |
596 | int padding) | |
597 | { | |
598 | int i, j, k, l, bit_alloc_bits, b, ch; | |
599 | unsigned char *sf; | |
600 | int q[3]; | |
601 | PutBitContext *p = &s->pb; | |
602 | ||
603 | /* header */ | |
604 | ||
605 | put_bits(p, 12, 0xfff); | |
606 | put_bits(p, 1, 1 - s->lsf); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */ | |
607 | put_bits(p, 2, 4-2); /* layer 2 */ | |
608 | put_bits(p, 1, 1); /* no error protection */ | |
609 | put_bits(p, 4, s->bitrate_index); | |
610 | put_bits(p, 2, s->freq_index); | |
611 | put_bits(p, 1, s->do_padding); /* use padding */ | |
612 | put_bits(p, 1, 0); /* private_bit */ | |
613 | put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO); | |
614 | put_bits(p, 2, 0); /* mode_ext */ | |
615 | put_bits(p, 1, 0); /* no copyright */ | |
616 | put_bits(p, 1, 1); /* original */ | |
617 | put_bits(p, 2, 0); /* no emphasis */ | |
618 | ||
619 | /* bit allocation */ | |
620 | j = 0; | |
621 | for(i=0;i<s->sblimit;i++) { | |
622 | bit_alloc_bits = s->alloc_table[j]; | |
623 | for(ch=0;ch<s->nb_channels;ch++) { | |
624 | put_bits(p, bit_alloc_bits, bit_alloc[ch][i]); | |
625 | } | |
626 | j += 1 << bit_alloc_bits; | |
627 | } | |
628 | ||
629 | /* scale codes */ | |
630 | for(i=0;i<s->sblimit;i++) { | |
631 | for(ch=0;ch<s->nb_channels;ch++) { | |
632 | if (bit_alloc[ch][i]) | |
633 | put_bits(p, 2, s->scale_code[ch][i]); | |
634 | } | |
635 | } | |
636 | ||
637 | /* scale factors */ | |
638 | for(i=0;i<s->sblimit;i++) { | |
639 | for(ch=0;ch<s->nb_channels;ch++) { | |
640 | if (bit_alloc[ch][i]) { | |
641 | sf = &s->scale_factors[ch][i][0]; | |
642 | switch(s->scale_code[ch][i]) { | |
643 | case 0: | |
644 | put_bits(p, 6, sf[0]); | |
645 | put_bits(p, 6, sf[1]); | |
646 | put_bits(p, 6, sf[2]); | |
647 | break; | |
648 | case 3: | |
649 | case 1: | |
650 | put_bits(p, 6, sf[0]); | |
651 | put_bits(p, 6, sf[2]); | |
652 | break; | |
653 | case 2: | |
654 | put_bits(p, 6, sf[0]); | |
655 | break; | |
656 | } | |
657 | } | |
658 | } | |
659 | } | |
660 | ||
661 | /* quantization & write sub band samples */ | |
662 | ||
663 | for(k=0;k<3;k++) { | |
664 | for(l=0;l<12;l+=3) { | |
665 | j = 0; | |
666 | for(i=0;i<s->sblimit;i++) { | |
667 | bit_alloc_bits = s->alloc_table[j]; | |
668 | for(ch=0;ch<s->nb_channels;ch++) { | |
669 | b = bit_alloc[ch][i]; | |
670 | if (b) { | |
671 | int qindex, steps, m, sample, bits; | |
672 | /* we encode 3 sub band samples of the same sub band at a time */ | |
673 | qindex = s->alloc_table[j+b]; | |
674 | steps = quant_steps[qindex]; | |
675 | for(m=0;m<3;m++) { | |
676 | sample = s->sb_samples[ch][k][l + m][i]; | |
677 | /* divide by scale factor */ | |
678 | #ifdef USE_FLOATS | |
679 | { | |
680 | float a; | |
681 | a = (float)sample * scale_factor_inv_table[s->scale_factors[ch][i][k]]; | |
682 | q[m] = (int)((a + 1.0) * steps * 0.5); | |
683 | } | |
684 | #else | |
685 | { | |
686 | int q1, e, shift, mult; | |
687 | e = s->scale_factors[ch][i][k]; | |
688 | shift = scale_factor_shift[e]; | |
689 | mult = scale_factor_mult[e]; | |
690 | ||
691 | /* normalize to P bits */ | |
692 | if (shift < 0) | |
693 | q1 = sample << (-shift); | |
694 | else | |
695 | q1 = sample >> shift; | |
696 | q1 = (q1 * mult) >> P; | |
697 | q[m] = ((q1 + (1 << P)) * steps) >> (P + 1); | |
698 | } | |
699 | #endif | |
700 | if (q[m] >= steps) | |
701 | q[m] = steps - 1; | |
702 | assert(q[m] >= 0 && q[m] < steps); | |
703 | } | |
704 | bits = quant_bits[qindex]; | |
705 | if (bits < 0) { | |
706 | /* group the 3 values to save bits */ | |
707 | put_bits(p, -bits, | |
708 | q[0] + steps * (q[1] + steps * q[2])); | |
709 | #if 0 | |
710 | printf("%d: gr1 %d\n", | |
711 | i, q[0] + steps * (q[1] + steps * q[2])); | |
712 | #endif | |
713 | } else { | |
714 | #if 0 | |
715 | printf("%d: gr3 %d %d %d\n", | |
716 | i, q[0], q[1], q[2]); | |
717 | #endif | |
718 | put_bits(p, bits, q[0]); | |
719 | put_bits(p, bits, q[1]); | |
720 | put_bits(p, bits, q[2]); | |
721 | } | |
722 | } | |
723 | } | |
724 | /* next subband in alloc table */ | |
725 | j += 1 << bit_alloc_bits; | |
726 | } | |
727 | } | |
728 | } | |
729 | ||
730 | /* padding */ | |
731 | for(i=0;i<padding;i++) | |
732 | put_bits(p, 1, 0); | |
733 | ||
734 | /* flush */ | |
735 | flush_put_bits(p); | |
736 | } | |
737 | ||
738 | int MPA_encode_frame(AVCodecContext *avctx, | |
739 | unsigned char *frame, int buf_size, void *data) | |
740 | { | |
741 | MpegAudioContext *s = avctx->priv_data; | |
742 | short *samples = data; | |
743 | short smr[MPA_MAX_CHANNELS][SBLIMIT]; | |
744 | unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT]; | |
745 | int padding, i; | |
746 | ||
747 | for(i=0;i<s->nb_channels;i++) { | |
748 | filter(s, i, samples + i, s->nb_channels); | |
749 | } | |
750 | ||
751 | for(i=0;i<s->nb_channels;i++) { | |
752 | compute_scale_factors(s->scale_code[i], s->scale_factors[i], | |
753 | s->sb_samples[i], s->sblimit); | |
754 | } | |
755 | for(i=0;i<s->nb_channels;i++) { | |
756 | psycho_acoustic_model(s, smr[i]); | |
757 | } | |
758 | compute_bit_allocation(s, smr, bit_alloc, &padding); | |
759 | ||
760 | init_put_bits(&s->pb, frame, MPA_MAX_CODED_FRAME_SIZE, NULL, NULL); | |
761 | ||
762 | encode_frame(s, bit_alloc, padding); | |
763 | ||
764 | s->nb_samples += MPA_FRAME_SIZE; | |
17592475 | 765 | return pbBufPtr(&s->pb) - s->pb.buf; |
de6d9b64 FB |
766 | } |
767 | ||
768 | ||
769 | AVCodec mp2_encoder = { | |
770 | "mp2", | |
771 | CODEC_TYPE_AUDIO, | |
772 | CODEC_ID_MP2, | |
773 | sizeof(MpegAudioContext), | |
774 | MPA_encode_init, | |
775 | MPA_encode_frame, | |
776 | NULL, | |
777 | }; | |
cd4af68a ZK |
778 | |
779 | #undef FIX |