/src/ffmpeg/libavcodec/mpegaudioenc.c
Line | Count | Source |
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 "config.h" |
28 | | #include "config_components.h" |
29 | | |
30 | | #include "libavutil/avassert.h" |
31 | | #include "libavutil/channel_layout.h" |
32 | | |
33 | | #include "avcodec.h" |
34 | | #include "codec_internal.h" |
35 | | #include "encode.h" |
36 | | #include "put_bits.h" |
37 | | |
38 | 0 | #define FRAC_BITS 15 /* fractional bits for sb_samples and dct */ |
39 | 0 | #define WFRAC_BITS 14 /* fractional bits for window */ |
40 | | |
41 | | #include "mpegaudio.h" |
42 | | #include "mpegaudiodsp.h" |
43 | | #include "mpegaudiodata.h" |
44 | | #include "mpegaudiotab.h" |
45 | | |
46 | | /* currently, cannot change these constants (need to modify |
47 | | quantization stage) */ |
48 | 0 | #define MUL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS) |
49 | | |
50 | 0 | #define SAMPLES_BUF_SIZE 4096 |
51 | | |
52 | | typedef struct MpegAudioContext { |
53 | | int nb_channels; |
54 | | int lsf; /* 1 if mpeg2 low bitrate selected */ |
55 | | int bitrate_index; /* bit rate */ |
56 | | int freq_index; |
57 | | int frame_size; /* frame size, in bits, without padding */ |
58 | | int is_fixed; |
59 | | /* padding computation */ |
60 | | int frame_frac, frame_frac_incr, do_padding; |
61 | | short samples_buf[MPA_MAX_CHANNELS][SAMPLES_BUF_SIZE]; /* buffer for filter */ |
62 | | int samples_offset[MPA_MAX_CHANNELS]; /* offset in samples_buf */ |
63 | | int sb_samples[MPA_MAX_CHANNELS][3][12][SBLIMIT]; |
64 | | unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3]; /* scale factors */ |
65 | | /* code to group 3 scale factors */ |
66 | | unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT]; |
67 | | int sblimit; /* number of used subbands */ |
68 | | const unsigned char *alloc_table; |
69 | | int16_t filter_bank[512]; |
70 | | int scale_factor_table[64]; |
71 | | unsigned char scale_diff_table[128]; |
72 | | union { |
73 | | float scale_factor_inv_table[64]; |
74 | | struct { |
75 | | int8_t scale_factor_shift[64]; |
76 | | unsigned short scale_factor_mult[64]; |
77 | | }; |
78 | | }; |
79 | | unsigned short total_quant_bits[17]; /* total number of bits per allocation group */ |
80 | | } MpegAudioContext; |
81 | | |
82 | 0 | #define IS_FIXED(s) (CONFIG_MP2_ENCODER && CONFIG_MP2FIXED_ENCODER ? (s)->is_fixed : CONFIG_MP2FIXED_ENCODER) |
83 | | |
84 | | static av_cold int mpa_encode_init(AVCodecContext *avctx) |
85 | 0 | { |
86 | 0 | MpegAudioContext *s = avctx->priv_data; |
87 | 0 | int freq = avctx->sample_rate; |
88 | 0 | int bitrate = avctx->bit_rate; |
89 | 0 | int channels = avctx->ch_layout.nb_channels; |
90 | 0 | int i, v, table; |
91 | 0 | float a; |
92 | |
|
93 | 0 | bitrate = bitrate / 1000; |
94 | 0 | s->nb_channels = channels; |
95 | 0 | avctx->frame_size = MPA_FRAME_SIZE; |
96 | 0 | avctx->initial_padding = 512 - 32 + 1; |
97 | | |
98 | | /* encoding freq */ |
99 | 0 | s->lsf = 0; |
100 | 0 | for (i = 0;; i++) { |
101 | 0 | av_assert1(i < 3); |
102 | 0 | if (ff_mpa_freq_tab[i] == freq) |
103 | 0 | break; |
104 | 0 | if ((ff_mpa_freq_tab[i] / 2) == freq) { |
105 | 0 | s->lsf = 1; |
106 | 0 | break; |
107 | 0 | } |
108 | 0 | } |
109 | 0 | s->freq_index = i; |
110 | | |
111 | | /* encoding bitrate & frequency */ |
112 | 0 | for(i=1;i<15;i++) { |
113 | 0 | if (ff_mpa_bitrate_tab[s->lsf][1][i] == bitrate) |
114 | 0 | break; |
115 | 0 | } |
116 | 0 | if (i == 15 && !avctx->bit_rate) { |
117 | 0 | i = 14; |
118 | 0 | bitrate = ff_mpa_bitrate_tab[s->lsf][1][i]; |
119 | 0 | avctx->bit_rate = bitrate * 1000; |
120 | 0 | } |
121 | 0 | if (i == 15){ |
122 | 0 | av_log(avctx, AV_LOG_ERROR, "bitrate %d is not allowed in mp2\n", bitrate); |
123 | 0 | return AVERROR(EINVAL); |
124 | 0 | } |
125 | 0 | s->bitrate_index = i; |
126 | | |
127 | | /* compute total header size & pad bit */ |
128 | |
|
129 | 0 | a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0); |
130 | 0 | s->frame_size = ((int)a) * 8; |
131 | | |
132 | | /* frame fractional size to compute padding */ |
133 | 0 | s->frame_frac = 0; |
134 | 0 | s->frame_frac_incr = (int)((a - floor(a)) * 65536.0); |
135 | | |
136 | | /* select the right allocation table */ |
137 | 0 | table = ff_mpa_l2_select_table(bitrate, s->nb_channels, freq, s->lsf); |
138 | | |
139 | | /* number of used subbands */ |
140 | 0 | s->sblimit = ff_mpa_sblimit_table[table]; |
141 | 0 | s->alloc_table = ff_mpa_alloc_tables[table]; |
142 | |
|
143 | 0 | ff_dlog(avctx, "%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n", |
144 | 0 | bitrate, freq, s->frame_size, table, s->frame_frac_incr); |
145 | |
|
146 | 0 | for(i=0;i<s->nb_channels;i++) |
147 | 0 | s->samples_offset[i] = 0; |
148 | |
|
149 | 0 | for(i=0;i<257;i++) { |
150 | 0 | int v; |
151 | 0 | v = ff_mpa_enwindow[i]; |
152 | 0 | #if WFRAC_BITS != 16 |
153 | 0 | v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS); |
154 | 0 | #endif |
155 | 0 | s->filter_bank[i] = v; |
156 | 0 | if ((i & 63) != 0) |
157 | 0 | v = -v; |
158 | 0 | if (i != 0) |
159 | 0 | s->filter_bank[512 - i] = v; |
160 | 0 | } |
161 | |
|
162 | 0 | for(i=0;i<64;i++) { |
163 | 0 | v = (int)(exp2((3 - i) / 3.0) * (1 << 20)); |
164 | 0 | if (v <= 0) |
165 | 0 | v = 1; |
166 | 0 | s->scale_factor_table[i] = v; |
167 | 0 | if (IS_FIXED(s)) { |
168 | 0 | #define P 15 |
169 | 0 | s->scale_factor_shift[i] = 21 - P - (i / 3); |
170 | 0 | s->scale_factor_mult[i] = (1 << P) * exp2((i % 3) / 3.0); |
171 | 0 | } else { |
172 | 0 | s->scale_factor_inv_table[i] = exp2(-(3 - i) / 3.0) / (float)(1 << 20); |
173 | 0 | } |
174 | 0 | } |
175 | 0 | for(i=0;i<128;i++) { |
176 | 0 | v = i - 64; |
177 | 0 | if (v <= -3) |
178 | 0 | v = 0; |
179 | 0 | else if (v < 0) |
180 | 0 | v = 1; |
181 | 0 | else if (v == 0) |
182 | 0 | v = 2; |
183 | 0 | else if (v < 3) |
184 | 0 | v = 3; |
185 | 0 | else |
186 | 0 | v = 4; |
187 | 0 | s->scale_diff_table[i] = v; |
188 | 0 | } |
189 | |
|
190 | 0 | for(i=0;i<17;i++) { |
191 | 0 | v = ff_mpa_quant_bits[i]; |
192 | 0 | if (v < 0) |
193 | 0 | v = -v; |
194 | 0 | else |
195 | 0 | v = v * 3; |
196 | 0 | s->total_quant_bits[i] = 12 * v; |
197 | 0 | } |
198 | |
|
199 | 0 | return 0; |
200 | 0 | } |
201 | | |
202 | | /* 32 point floating point IDCT without 1/sqrt(2) coef zero scaling */ |
203 | | static void idct32(int *out, int *tab) |
204 | 0 | { |
205 | 0 | int i, j; |
206 | 0 | int *t, *t1, xr; |
207 | 0 | const int *xp = costab32; |
208 | |
|
209 | 0 | for(j=31;j>=3;j-=2) tab[j] += tab[j - 2]; |
210 | |
|
211 | 0 | t = tab + 30; |
212 | 0 | t1 = tab + 2; |
213 | 0 | do { |
214 | 0 | t[0] += t[-4]; |
215 | 0 | t[1] += t[1 - 4]; |
216 | 0 | t -= 4; |
217 | 0 | } while (t != t1); |
218 | |
|
219 | 0 | t = tab + 28; |
220 | 0 | t1 = tab + 4; |
221 | 0 | do { |
222 | 0 | t[0] += t[-8]; |
223 | 0 | t[1] += t[1-8]; |
224 | 0 | t[2] += t[2-8]; |
225 | 0 | t[3] += t[3-8]; |
226 | 0 | t -= 8; |
227 | 0 | } while (t != t1); |
228 | |
|
229 | 0 | t = tab; |
230 | 0 | t1 = tab + 32; |
231 | 0 | do { |
232 | 0 | t[ 3] = -t[ 3]; |
233 | 0 | t[ 6] = -t[ 6]; |
234 | |
|
235 | 0 | t[11] = -t[11]; |
236 | 0 | t[12] = -t[12]; |
237 | 0 | t[13] = -t[13]; |
238 | 0 | t[15] = -t[15]; |
239 | 0 | t += 16; |
240 | 0 | } while (t != t1); |
241 | | |
242 | |
|
243 | 0 | t = tab; |
244 | 0 | t1 = tab + 8; |
245 | 0 | do { |
246 | 0 | int x1, x2, x3, x4; |
247 | |
|
248 | 0 | x3 = MUL(t[16], FIX(M_SQRT2*0.5)); |
249 | 0 | x4 = t[0] - x3; |
250 | 0 | x3 = t[0] + x3; |
251 | |
|
252 | 0 | x2 = MUL(-(t[24] + t[8]), FIX(M_SQRT2*0.5)); |
253 | 0 | x1 = MUL((t[8] - x2), xp[0]); |
254 | 0 | x2 = MUL((t[8] + x2), xp[1]); |
255 | |
|
256 | 0 | t[ 0] = x3 + x1; |
257 | 0 | t[ 8] = x4 - x2; |
258 | 0 | t[16] = x4 + x2; |
259 | 0 | t[24] = x3 - x1; |
260 | 0 | t++; |
261 | 0 | } while (t != t1); |
262 | |
|
263 | 0 | xp += 2; |
264 | 0 | t = tab; |
265 | 0 | t1 = tab + 4; |
266 | 0 | do { |
267 | 0 | xr = MUL(t[28],xp[0]); |
268 | 0 | t[28] = (t[0] - xr); |
269 | 0 | t[0] = (t[0] + xr); |
270 | |
|
271 | 0 | xr = MUL(t[4],xp[1]); |
272 | 0 | t[ 4] = (t[24] - xr); |
273 | 0 | t[24] = (t[24] + xr); |
274 | |
|
275 | 0 | xr = MUL(t[20],xp[2]); |
276 | 0 | t[20] = (t[8] - xr); |
277 | 0 | t[ 8] = (t[8] + xr); |
278 | |
|
279 | 0 | xr = MUL(t[12],xp[3]); |
280 | 0 | t[12] = (t[16] - xr); |
281 | 0 | t[16] = (t[16] + xr); |
282 | 0 | t++; |
283 | 0 | } while (t != t1); |
284 | 0 | xp += 4; |
285 | |
|
286 | 0 | for (i = 0; i < 4; i++) { |
287 | 0 | xr = MUL(tab[30-i*4],xp[0]); |
288 | 0 | tab[30-i*4] = (tab[i*4] - xr); |
289 | 0 | tab[ i*4] = (tab[i*4] + xr); |
290 | |
|
291 | 0 | xr = MUL(tab[ 2+i*4],xp[1]); |
292 | 0 | tab[ 2+i*4] = (tab[28-i*4] - xr); |
293 | 0 | tab[28-i*4] = (tab[28-i*4] + xr); |
294 | |
|
295 | 0 | xr = MUL(tab[31-i*4],xp[0]); |
296 | 0 | tab[31-i*4] = (tab[1+i*4] - xr); |
297 | 0 | tab[ 1+i*4] = (tab[1+i*4] + xr); |
298 | |
|
299 | 0 | xr = MUL(tab[ 3+i*4],xp[1]); |
300 | 0 | tab[ 3+i*4] = (tab[29-i*4] - xr); |
301 | 0 | tab[29-i*4] = (tab[29-i*4] + xr); |
302 | |
|
303 | 0 | xp += 2; |
304 | 0 | } |
305 | |
|
306 | 0 | t = tab + 30; |
307 | 0 | t1 = tab + 1; |
308 | 0 | do { |
309 | 0 | xr = MUL(t1[0], *xp); |
310 | 0 | t1[0] = (t[0] - xr); |
311 | 0 | t[0] = (t[0] + xr); |
312 | 0 | t -= 2; |
313 | 0 | t1 += 2; |
314 | 0 | xp++; |
315 | 0 | } while (t >= tab); |
316 | |
|
317 | 0 | for(i=0;i<32;i++) { |
318 | 0 | out[i] = tab[bitinv32[i]]; |
319 | 0 | } |
320 | 0 | } |
321 | | |
322 | 0 | #define WSHIFT (WFRAC_BITS + 15 - FRAC_BITS) |
323 | | |
324 | | static void filter(MpegAudioContext *s, int ch, const short *samples, int incr) |
325 | 0 | { |
326 | 0 | short *p, *q; |
327 | 0 | int sum, offset, i, j; |
328 | 0 | int tmp[64]; |
329 | 0 | int tmp1[32]; |
330 | 0 | int *out; |
331 | |
|
332 | 0 | offset = s->samples_offset[ch]; |
333 | 0 | out = &s->sb_samples[ch][0][0][0]; |
334 | 0 | for(j=0;j<36;j++) { |
335 | | /* 32 samples at once */ |
336 | 0 | for(i=0;i<32;i++) { |
337 | 0 | s->samples_buf[ch][offset + (31 - i)] = samples[0]; |
338 | 0 | samples += incr; |
339 | 0 | } |
340 | | |
341 | | /* filter */ |
342 | 0 | p = s->samples_buf[ch] + offset; |
343 | 0 | q = s->filter_bank; |
344 | | /* maxsum = 23169 */ |
345 | 0 | for(i=0;i<64;i++) { |
346 | 0 | sum = p[0*64] * q[0*64]; |
347 | 0 | sum += p[1*64] * q[1*64]; |
348 | 0 | sum += p[2*64] * q[2*64]; |
349 | 0 | sum += p[3*64] * q[3*64]; |
350 | 0 | sum += p[4*64] * q[4*64]; |
351 | 0 | sum += p[5*64] * q[5*64]; |
352 | 0 | sum += p[6*64] * q[6*64]; |
353 | 0 | sum += p[7*64] * q[7*64]; |
354 | 0 | tmp[i] = sum; |
355 | 0 | p++; |
356 | 0 | q++; |
357 | 0 | } |
358 | 0 | tmp1[0] = tmp[16] >> WSHIFT; |
359 | 0 | for( i=1; i<=16; i++ ) tmp1[i] = (tmp[i+16]+tmp[16-i]) >> WSHIFT; |
360 | 0 | for( i=17; i<=31; i++ ) tmp1[i] = (tmp[i+16]-tmp[80-i]) >> WSHIFT; |
361 | |
|
362 | 0 | idct32(out, tmp1); |
363 | | |
364 | | /* advance of 32 samples */ |
365 | 0 | offset -= 32; |
366 | 0 | out += 32; |
367 | | /* handle the wrap around */ |
368 | 0 | if (offset < 0) { |
369 | 0 | memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32), |
370 | 0 | s->samples_buf[ch], (512 - 32) * 2); |
371 | 0 | offset = SAMPLES_BUF_SIZE - 512; |
372 | 0 | } |
373 | 0 | } |
374 | 0 | s->samples_offset[ch] = offset; |
375 | 0 | } |
376 | | |
377 | | static void compute_scale_factors(MpegAudioContext *s, |
378 | | unsigned char scale_code[SBLIMIT], |
379 | | unsigned char scale_factors[SBLIMIT][3], |
380 | | int sb_samples[3][12][SBLIMIT], |
381 | | int sblimit) |
382 | 0 | { |
383 | 0 | int *p, vmax, v, n, i, j, k, code; |
384 | 0 | int index, d1, d2; |
385 | 0 | unsigned char *sf = &scale_factors[0][0]; |
386 | |
|
387 | 0 | for(j=0;j<sblimit;j++) { |
388 | 0 | for(i=0;i<3;i++) { |
389 | | /* find the max absolute value */ |
390 | 0 | p = &sb_samples[i][0][j]; |
391 | 0 | vmax = abs(*p); |
392 | 0 | for(k=1;k<12;k++) { |
393 | 0 | p += SBLIMIT; |
394 | 0 | v = abs(*p); |
395 | 0 | if (v > vmax) |
396 | 0 | vmax = v; |
397 | 0 | } |
398 | | /* compute the scale factor index using log 2 computations */ |
399 | 0 | if (vmax > 1) { |
400 | 0 | n = av_log2(vmax); |
401 | | /* n is the position of the MSB of vmax. now |
402 | | use at most 2 compares to find the index */ |
403 | 0 | index = (21 - n) * 3 - 3; |
404 | 0 | if (index >= 0) { |
405 | 0 | while (vmax <= s->scale_factor_table[index+1]) |
406 | 0 | index++; |
407 | 0 | } else { |
408 | 0 | index = 0; /* very unlikely case of overflow */ |
409 | 0 | } |
410 | 0 | } else { |
411 | 0 | index = 62; /* value 63 is not allowed */ |
412 | 0 | } |
413 | |
|
414 | 0 | ff_dlog(NULL, "%2d:%d in=%x %x %d\n", |
415 | 0 | j, i, vmax, s->scale_factor_table[index], index); |
416 | | /* store the scale factor */ |
417 | 0 | av_assert2(index >=0 && index <= 63); |
418 | 0 | sf[i] = index; |
419 | 0 | } |
420 | | |
421 | | /* compute the transmission factor : look if the scale factors |
422 | | are close enough to each other */ |
423 | 0 | d1 = s->scale_diff_table[sf[0] - sf[1] + 64]; |
424 | 0 | d2 = s->scale_diff_table[sf[1] - sf[2] + 64]; |
425 | | |
426 | | /* handle the 25 cases */ |
427 | 0 | switch(d1 * 5 + d2) { |
428 | 0 | case 0*5+0: |
429 | 0 | case 0*5+4: |
430 | 0 | case 3*5+4: |
431 | 0 | case 4*5+0: |
432 | 0 | case 4*5+4: |
433 | 0 | code = 0; |
434 | 0 | break; |
435 | 0 | case 0*5+1: |
436 | 0 | case 0*5+2: |
437 | 0 | case 4*5+1: |
438 | 0 | case 4*5+2: |
439 | 0 | code = 3; |
440 | 0 | sf[2] = sf[1]; |
441 | 0 | break; |
442 | 0 | case 0*5+3: |
443 | 0 | case 4*5+3: |
444 | 0 | code = 3; |
445 | 0 | sf[1] = sf[2]; |
446 | 0 | break; |
447 | 0 | case 1*5+0: |
448 | 0 | case 1*5+4: |
449 | 0 | case 2*5+4: |
450 | 0 | code = 1; |
451 | 0 | sf[1] = sf[0]; |
452 | 0 | break; |
453 | 0 | case 1*5+1: |
454 | 0 | case 1*5+2: |
455 | 0 | case 2*5+0: |
456 | 0 | case 2*5+1: |
457 | 0 | case 2*5+2: |
458 | 0 | code = 2; |
459 | 0 | sf[1] = sf[2] = sf[0]; |
460 | 0 | break; |
461 | 0 | case 2*5+3: |
462 | 0 | case 3*5+3: |
463 | 0 | code = 2; |
464 | 0 | sf[0] = sf[1] = sf[2]; |
465 | 0 | break; |
466 | 0 | case 3*5+0: |
467 | 0 | case 3*5+1: |
468 | 0 | case 3*5+2: |
469 | 0 | code = 2; |
470 | 0 | sf[0] = sf[2] = sf[1]; |
471 | 0 | break; |
472 | 0 | case 1*5+3: |
473 | 0 | code = 2; |
474 | 0 | if (sf[0] > sf[2]) |
475 | 0 | sf[0] = sf[2]; |
476 | 0 | sf[1] = sf[2] = sf[0]; |
477 | 0 | break; |
478 | 0 | default: |
479 | 0 | av_assert2(0); //cannot happen |
480 | 0 | code = 0; /* kill warning */ |
481 | 0 | } |
482 | | |
483 | 0 | ff_dlog(NULL, "%d: %2d %2d %2d %d %d -> %d\n", j, |
484 | 0 | sf[0], sf[1], sf[2], d1, d2, code); |
485 | 0 | scale_code[j] = code; |
486 | 0 | sf += 3; |
487 | 0 | } |
488 | 0 | } |
489 | | |
490 | | /* The most important function : psycho acoustic module. In this |
491 | | encoder there is basically none, so this is the worst you can do, |
492 | | but also this is the simpler. */ |
493 | | static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT]) |
494 | 0 | { |
495 | 0 | int i; |
496 | |
|
497 | 0 | for(i=0;i<s->sblimit;i++) { |
498 | 0 | smr[i] = (int)(fixed_smr[i] * 10); |
499 | 0 | } |
500 | 0 | } |
501 | | |
502 | | |
503 | 0 | #define SB_NOTALLOCATED 0 |
504 | 0 | #define SB_ALLOCATED 1 |
505 | 0 | #define SB_NOMORE 2 |
506 | | |
507 | | /* Try to maximize the smr while using a number of bits inferior to |
508 | | the frame size. I tried to make the code simpler, faster and |
509 | | smaller than other encoders :-) */ |
510 | | static unsigned compute_bit_allocation(MpegAudioContext *s, |
511 | | short smr1[MPA_MAX_CHANNELS][SBLIMIT], |
512 | | unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], |
513 | | int *padding) |
514 | 0 | { |
515 | 0 | int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size; |
516 | 0 | int incr; |
517 | 0 | short smr[MPA_MAX_CHANNELS][SBLIMIT]; |
518 | 0 | unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT]; |
519 | 0 | const unsigned char *alloc; |
520 | |
|
521 | 0 | memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT); |
522 | 0 | memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT); |
523 | 0 | memset(bit_alloc, 0, s->nb_channels * SBLIMIT); |
524 | | |
525 | | /* compute frame size and padding */ |
526 | 0 | max_frame_size = s->frame_size; |
527 | 0 | s->frame_frac += s->frame_frac_incr; |
528 | 0 | if (s->frame_frac >= 65536) { |
529 | 0 | s->frame_frac -= 65536; |
530 | 0 | s->do_padding = 1; |
531 | 0 | max_frame_size += 8; |
532 | 0 | } else { |
533 | 0 | s->do_padding = 0; |
534 | 0 | } |
535 | | |
536 | | /* compute the header + bit alloc size */ |
537 | 0 | current_frame_size = 32; |
538 | 0 | alloc = s->alloc_table; |
539 | 0 | for(i=0;i<s->sblimit;i++) { |
540 | 0 | incr = alloc[0]; |
541 | 0 | current_frame_size += incr * s->nb_channels; |
542 | 0 | alloc += 1 << incr; |
543 | 0 | } |
544 | 0 | for(;;) { |
545 | | /* look for the subband with the largest signal to mask ratio */ |
546 | 0 | max_sb = -1; |
547 | 0 | max_ch = -1; |
548 | 0 | max_smr = INT_MIN; |
549 | 0 | for(ch=0;ch<s->nb_channels;ch++) { |
550 | 0 | for(i=0;i<s->sblimit;i++) { |
551 | 0 | if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) { |
552 | 0 | max_smr = smr[ch][i]; |
553 | 0 | max_sb = i; |
554 | 0 | max_ch = ch; |
555 | 0 | } |
556 | 0 | } |
557 | 0 | } |
558 | 0 | if (max_sb < 0) |
559 | 0 | break; |
560 | 0 | ff_dlog(NULL, "current=%d max=%d max_sb=%d max_ch=%d alloc=%d\n", |
561 | 0 | current_frame_size, max_frame_size, max_sb, max_ch, |
562 | 0 | bit_alloc[max_ch][max_sb]); |
563 | | |
564 | | /* find alloc table entry (XXX: not optimal, should use |
565 | | pointer table) */ |
566 | 0 | alloc = s->alloc_table; |
567 | 0 | for(i=0;i<max_sb;i++) { |
568 | 0 | alloc += 1 << alloc[0]; |
569 | 0 | } |
570 | |
|
571 | 0 | if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) { |
572 | | /* nothing was coded for this band: add the necessary bits */ |
573 | 0 | incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6; |
574 | 0 | incr += s->total_quant_bits[alloc[1]]; |
575 | 0 | } else { |
576 | | /* increments bit allocation */ |
577 | 0 | b = bit_alloc[max_ch][max_sb]; |
578 | 0 | incr = s->total_quant_bits[alloc[b + 1]] - |
579 | 0 | s->total_quant_bits[alloc[b]]; |
580 | 0 | } |
581 | |
|
582 | 0 | if (current_frame_size + incr <= max_frame_size) { |
583 | | /* can increase size */ |
584 | 0 | b = ++bit_alloc[max_ch][max_sb]; |
585 | 0 | current_frame_size += incr; |
586 | | /* decrease smr by the resolution we added */ |
587 | 0 | smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]]; |
588 | | /* max allocation size reached ? */ |
589 | 0 | if (b == ((1 << alloc[0]) - 1)) |
590 | 0 | subband_status[max_ch][max_sb] = SB_NOMORE; |
591 | 0 | else |
592 | 0 | subband_status[max_ch][max_sb] = SB_ALLOCATED; |
593 | 0 | } else { |
594 | | /* cannot increase the size of this subband */ |
595 | 0 | subband_status[max_ch][max_sb] = SB_NOMORE; |
596 | 0 | } |
597 | 0 | } |
598 | 0 | *padding = max_frame_size - current_frame_size; |
599 | 0 | av_assert0(*padding >= 0); |
600 | 0 | return max_frame_size / 8U; |
601 | 0 | } |
602 | | |
603 | | /// Quantization & write sub band samples |
604 | | static av_always_inline void encode_subbands(MpegAudioContext *const s, |
605 | | PutBitContext *const p, |
606 | | const uint8_t bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], |
607 | | int is_fixed) |
608 | 0 | { |
609 | 0 | for (int k = 0; k < 3; ++k) { |
610 | 0 | for (int l = 0; l < 12; l += 3) { |
611 | 0 | for (int i = 0, j = 0; i < s->sblimit; ++i) { |
612 | 0 | const int bit_alloc_bits = s->alloc_table[j]; |
613 | 0 | for (int ch = 0; ch < s->nb_channels; ++ch) { |
614 | 0 | const int b = bit_alloc[ch][i]; |
615 | 0 | if (b) { |
616 | | /* we encode 3 sub band samples of the same sub band at a time */ |
617 | 0 | const int qindex = s->alloc_table[j + b]; |
618 | 0 | const int steps = ff_mpa_quant_steps[qindex]; |
619 | 0 | int q[3]; |
620 | |
|
621 | 0 | for (int m = 0; m < 3; ++m) { |
622 | 0 | const int sample = s->sb_samples[ch][k][l + m][i]; |
623 | | /* divide by scale factor */ |
624 | 0 | if (!is_fixed) { |
625 | 0 | float a = (float)sample * s->scale_factor_inv_table[s->scale_factors[ch][i][k]]; |
626 | 0 | q[m] = (int)((a + 1.0) * steps * 0.5); |
627 | 0 | } else { |
628 | 0 | const int e = s->scale_factors[ch][i][k]; |
629 | 0 | const int shift = s->scale_factor_shift[e]; |
630 | 0 | const int mult = s->scale_factor_mult[e]; |
631 | 0 | int q1; |
632 | | |
633 | | /* normalize to P bits */ |
634 | 0 | if (shift < 0) |
635 | 0 | q1 = sample * (1 << -shift); |
636 | 0 | else |
637 | 0 | q1 = sample >> shift; |
638 | 0 | q1 = (q1 * mult) >> P; |
639 | 0 | q1 += 1 << P; |
640 | 0 | if (q1 < 0) |
641 | 0 | q1 = 0; |
642 | 0 | q[m] = (q1 * (unsigned)steps) >> (P + 1); |
643 | 0 | } |
644 | 0 | if (q[m] >= steps) |
645 | 0 | q[m] = steps - 1; |
646 | 0 | av_assert2(q[m] >= 0 && q[m] < steps); |
647 | 0 | } |
648 | 0 | const int bits = ff_mpa_quant_bits[qindex]; |
649 | 0 | if (bits < 0) { |
650 | | /* group the 3 values to save bits */ |
651 | 0 | put_bits(p, -bits, |
652 | 0 | q[0] + steps * (q[1] + steps * q[2])); |
653 | 0 | } else { |
654 | 0 | put_bits(p, bits, q[0]); |
655 | 0 | put_bits(p, bits, q[1]); |
656 | 0 | put_bits(p, bits, q[2]); |
657 | 0 | } |
658 | 0 | } |
659 | 0 | } |
660 | | /* next subband in alloc table */ |
661 | 0 | j += 1 << bit_alloc_bits; |
662 | 0 | } |
663 | 0 | } |
664 | 0 | } |
665 | 0 | } |
666 | | |
667 | | /* |
668 | | * Output the MPEG audio layer 2 frame. Note how the code is small |
669 | | * compared to other encoders :-) |
670 | | */ |
671 | | static void encode_frame(MpegAudioContext *s, uint8_t *buf, unsigned buf_size, |
672 | | unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT], |
673 | | int padding) |
674 | 0 | { |
675 | 0 | int i, j, bit_alloc_bits, ch; |
676 | 0 | unsigned char *sf; |
677 | 0 | PutBitContext p0, *p = &p0; |
678 | |
|
679 | 0 | init_put_bits(p, buf, buf_size); |
680 | | |
681 | | /* header */ |
682 | |
|
683 | 0 | put_bits(p, 12, 0xfff); |
684 | 0 | put_bits(p, 1, 1 - s->lsf); /* 1 = MPEG-1 ID, 0 = MPEG-2 lsf ID */ |
685 | 0 | put_bits(p, 2, 4-2); /* layer 2 */ |
686 | 0 | put_bits(p, 1, 1); /* no error protection */ |
687 | 0 | put_bits(p, 4, s->bitrate_index); |
688 | 0 | put_bits(p, 2, s->freq_index); |
689 | 0 | put_bits(p, 1, s->do_padding); /* use padding */ |
690 | 0 | put_bits(p, 1, 0); /* private_bit */ |
691 | 0 | put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO); |
692 | 0 | put_bits(p, 2, 0); /* mode_ext */ |
693 | 0 | put_bits(p, 1, 0); /* no copyright */ |
694 | 0 | put_bits(p, 1, 1); /* original */ |
695 | 0 | put_bits(p, 2, 0); /* no emphasis */ |
696 | | |
697 | | /* bit allocation */ |
698 | 0 | j = 0; |
699 | 0 | for(i=0;i<s->sblimit;i++) { |
700 | 0 | bit_alloc_bits = s->alloc_table[j]; |
701 | 0 | for(ch=0;ch<s->nb_channels;ch++) { |
702 | 0 | put_bits(p, bit_alloc_bits, bit_alloc[ch][i]); |
703 | 0 | } |
704 | 0 | j += 1 << bit_alloc_bits; |
705 | 0 | } |
706 | | |
707 | | /* scale codes */ |
708 | 0 | for(i=0;i<s->sblimit;i++) { |
709 | 0 | for(ch=0;ch<s->nb_channels;ch++) { |
710 | 0 | if (bit_alloc[ch][i]) |
711 | 0 | put_bits(p, 2, s->scale_code[ch][i]); |
712 | 0 | } |
713 | 0 | } |
714 | | |
715 | | /* scale factors */ |
716 | 0 | for(i=0;i<s->sblimit;i++) { |
717 | 0 | for(ch=0;ch<s->nb_channels;ch++) { |
718 | 0 | if (bit_alloc[ch][i]) { |
719 | 0 | sf = &s->scale_factors[ch][i][0]; |
720 | 0 | switch(s->scale_code[ch][i]) { |
721 | 0 | case 0: |
722 | 0 | put_bits(p, 18, sf[0] << 12 | sf[1] << 6 | sf[2]); |
723 | 0 | break; |
724 | 0 | case 3: |
725 | 0 | case 1: |
726 | 0 | put_bits(p, 12, sf[0] << 6 | sf[2]); |
727 | 0 | break; |
728 | 0 | case 2: |
729 | 0 | put_bits(p, 6, sf[0]); |
730 | 0 | break; |
731 | 0 | } |
732 | 0 | } |
733 | 0 | } |
734 | 0 | } |
735 | | |
736 | | #if CONFIG_SMALL |
737 | | encode_subbands(s, p, bit_alloc, IS_FIXED(s)); |
738 | | #else |
739 | 0 | if (IS_FIXED(s)) |
740 | 0 | encode_subbands(s, p, bit_alloc, 1); |
741 | 0 | else |
742 | 0 | encode_subbands(s, p, bit_alloc, 0); |
743 | 0 | #endif |
744 | |
|
745 | 0 | av_assert1(put_bits_left(p) == padding); |
746 | | |
747 | | /* flush */ |
748 | 0 | flush_put_bits(p); |
749 | | |
750 | | /* padding */ |
751 | 0 | if (put_bytes_left(p, 0)) |
752 | 0 | memset(put_bits_ptr(p), 0, put_bytes_left(p, 0)); |
753 | 0 | } |
754 | | |
755 | | static int mpa_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, |
756 | | const AVFrame *frame, int *got_packet_ptr) |
757 | 0 | { |
758 | 0 | MpegAudioContext *s = avctx->priv_data; |
759 | 0 | const int16_t *samples = (const int16_t *)frame->data[0]; |
760 | 0 | short smr[MPA_MAX_CHANNELS][SBLIMIT]; |
761 | 0 | unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT]; |
762 | 0 | int padding, i, ret; |
763 | |
|
764 | 0 | for(i=0;i<s->nb_channels;i++) { |
765 | 0 | filter(s, i, samples + i, s->nb_channels); |
766 | 0 | } |
767 | |
|
768 | 0 | for(i=0;i<s->nb_channels;i++) { |
769 | 0 | compute_scale_factors(s, s->scale_code[i], s->scale_factors[i], |
770 | 0 | s->sb_samples[i], s->sblimit); |
771 | 0 | } |
772 | 0 | for(i=0;i<s->nb_channels;i++) { |
773 | 0 | psycho_acoustic_model(s, smr[i]); |
774 | 0 | } |
775 | 0 | unsigned frame_size = compute_bit_allocation(s, smr, bit_alloc, &padding); |
776 | |
|
777 | 0 | ret = ff_get_encode_buffer(avctx, avpkt, frame_size, 0); |
778 | 0 | if (ret < 0) |
779 | 0 | return ret; |
780 | | |
781 | 0 | encode_frame(s, avpkt->data, frame_size, bit_alloc, padding); |
782 | |
|
783 | 0 | if (frame->pts != AV_NOPTS_VALUE) |
784 | 0 | avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->initial_padding); |
785 | |
|
786 | 0 | *got_packet_ptr = 1; |
787 | 0 | return 0; |
788 | 0 | } |
789 | | |
790 | | static const FFCodecDefault mp2_defaults[] = { |
791 | | { "b", "0" }, |
792 | | { NULL }, |
793 | | }; |
794 | | |
795 | | #if CONFIG_MP2_ENCODER |
796 | | const FFCodec ff_mp2_encoder = { |
797 | | .p.name = "mp2", |
798 | | CODEC_LONG_NAME("MP2 (MPEG audio layer 2)"), |
799 | | .p.type = AVMEDIA_TYPE_AUDIO, |
800 | | .p.id = AV_CODEC_ID_MP2, |
801 | | .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_ENCODER_REORDERED_OPAQUE, |
802 | | .priv_data_size = sizeof(MpegAudioContext), |
803 | | .init = mpa_encode_init, |
804 | | FF_CODEC_ENCODE_CB(mpa_encode_frame), |
805 | | CODEC_SAMPLEFMTS(AV_SAMPLE_FMT_S16), |
806 | | CODEC_SAMPLERATES(44100, 48000, 32000, 22050, 24000, 16000), |
807 | | CODEC_CH_LAYOUTS(AV_CHANNEL_LAYOUT_MONO, AV_CHANNEL_LAYOUT_STEREO), |
808 | | .defaults = mp2_defaults, |
809 | | }; |
810 | | #endif |
811 | | |
812 | | #if CONFIG_MP2FIXED_ENCODER |
813 | | static av_cold int mpa_fixed_encode_init(AVCodecContext *avctx) |
814 | 0 | { |
815 | 0 | MpegAudioContext *s = avctx->priv_data; |
816 | |
|
817 | 0 | s->is_fixed = 1; |
818 | 0 | return mpa_encode_init(avctx); |
819 | 0 | } |
820 | | |
821 | | const FFCodec ff_mp2fixed_encoder = { |
822 | | .p.name = "mp2fixed", |
823 | | CODEC_LONG_NAME("MP2 fixed point (MPEG audio layer 2)"), |
824 | | .p.type = AVMEDIA_TYPE_AUDIO, |
825 | | .p.id = AV_CODEC_ID_MP2, |
826 | | .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_ENCODER_REORDERED_OPAQUE, |
827 | | .priv_data_size = sizeof(MpegAudioContext), |
828 | | .init = mpa_fixed_encode_init, |
829 | | FF_CODEC_ENCODE_CB(mpa_encode_frame), |
830 | | CODEC_SAMPLEFMTS(AV_SAMPLE_FMT_S16), |
831 | | CODEC_SAMPLERATES(44100, 48000, 32000, 22050, 24000, 16000), |
832 | | CODEC_CH_LAYOUTS(AV_CHANNEL_LAYOUT_MONO, AV_CHANNEL_LAYOUT_STEREO), |
833 | | .defaults = mp2_defaults, |
834 | | }; |
835 | | #endif |