bc99a443b3a0906542bb5786c604ce9a07aa31a5
[libav.git] / libavcodec / aacenc.c
1 /*
2 * AAC encoder
3 * Copyright (C) 2008 Konstantin Shishkov
4 *
5 * This file is part of Libav.
6 *
7 * Libav 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 * Libav 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 Libav; 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 * AAC encoder
25 */
26
27 /***********************************
28 * TODOs:
29 * add sane pulse detection
30 * add temporal noise shaping
31 ***********************************/
32
33 #include "libavutil/opt.h"
34 #include "avcodec.h"
35 #include "put_bits.h"
36 #include "dsputil.h"
37 #include "internal.h"
38 #include "mpeg4audio.h"
39 #include "kbdwin.h"
40 #include "sinewin.h"
41
42 #include "aac.h"
43 #include "aactab.h"
44 #include "aacenc.h"
45
46 #include "psymodel.h"
47
48 #define AAC_MAX_CHANNELS 6
49
50 #define ERROR_IF(cond, ...) \
51 if (cond) { \
52 av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \
53 return AVERROR(EINVAL); \
54 }
55
56 float ff_aac_pow34sf_tab[428];
57
58 static const uint8_t swb_size_1024_96[] = {
59 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8,
60 12, 12, 12, 12, 12, 16, 16, 24, 28, 36, 44,
61 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64
62 };
63
64 static const uint8_t swb_size_1024_64[] = {
65 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8,
66 12, 12, 12, 16, 16, 16, 20, 24, 24, 28, 36,
67 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40
68 };
69
70 static const uint8_t swb_size_1024_48[] = {
71 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8,
72 12, 12, 12, 12, 16, 16, 20, 20, 24, 24, 28, 28,
73 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
74 96
75 };
76
77 static const uint8_t swb_size_1024_32[] = {
78 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8,
79 12, 12, 12, 12, 16, 16, 20, 20, 24, 24, 28, 28,
80 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32
81 };
82
83 static const uint8_t swb_size_1024_24[] = {
84 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
85 12, 12, 12, 12, 16, 16, 16, 20, 20, 24, 24, 28, 28,
86 32, 36, 36, 40, 44, 48, 52, 52, 64, 64, 64, 64, 64
87 };
88
89 static const uint8_t swb_size_1024_16[] = {
90 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
91 12, 12, 12, 12, 12, 12, 12, 12, 12, 16, 16, 16, 16, 20, 20, 20, 24, 24, 28, 28,
92 32, 36, 40, 40, 44, 48, 52, 56, 60, 64, 64, 64
93 };
94
95 static const uint8_t swb_size_1024_8[] = {
96 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12,
97 16, 16, 16, 16, 16, 16, 16, 20, 20, 20, 20, 24, 24, 24, 28, 28,
98 32, 36, 36, 40, 44, 48, 52, 56, 60, 64, 80
99 };
100
101 static const uint8_t *swb_size_1024[] = {
102 swb_size_1024_96, swb_size_1024_96, swb_size_1024_64,
103 swb_size_1024_48, swb_size_1024_48, swb_size_1024_32,
104 swb_size_1024_24, swb_size_1024_24, swb_size_1024_16,
105 swb_size_1024_16, swb_size_1024_16, swb_size_1024_8
106 };
107
108 static const uint8_t swb_size_128_96[] = {
109 4, 4, 4, 4, 4, 4, 8, 8, 8, 16, 28, 36
110 };
111
112 static const uint8_t swb_size_128_48[] = {
113 4, 4, 4, 4, 4, 8, 8, 8, 12, 12, 12, 16, 16, 16
114 };
115
116 static const uint8_t swb_size_128_24[] = {
117 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 20
118 };
119
120 static const uint8_t swb_size_128_16[] = {
121 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 12, 12, 16, 20, 20
122 };
123
124 static const uint8_t swb_size_128_8[] = {
125 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 12, 16, 20, 20
126 };
127
128 static const uint8_t *swb_size_128[] = {
129 /* the last entry on the following row is swb_size_128_64 but is a
130 duplicate of swb_size_128_96 */
131 swb_size_128_96, swb_size_128_96, swb_size_128_96,
132 swb_size_128_48, swb_size_128_48, swb_size_128_48,
133 swb_size_128_24, swb_size_128_24, swb_size_128_16,
134 swb_size_128_16, swb_size_128_16, swb_size_128_8
135 };
136
137 /** default channel configurations */
138 static const uint8_t aac_chan_configs[6][5] = {
139 {1, TYPE_SCE}, // 1 channel - single channel element
140 {1, TYPE_CPE}, // 2 channels - channel pair
141 {2, TYPE_SCE, TYPE_CPE}, // 3 channels - center + stereo
142 {3, TYPE_SCE, TYPE_CPE, TYPE_SCE}, // 4 channels - front center + stereo + back center
143 {3, TYPE_SCE, TYPE_CPE, TYPE_CPE}, // 5 channels - front center + stereo + back stereo
144 {4, TYPE_SCE, TYPE_CPE, TYPE_CPE, TYPE_LFE}, // 6 channels - front center + stereo + back stereo + LFE
145 };
146
147 /**
148 * Table to remap channels from Libav's default order to AAC order.
149 */
150 static const uint8_t aac_chan_maps[AAC_MAX_CHANNELS][AAC_MAX_CHANNELS] = {
151 { 0 },
152 { 0, 1 },
153 { 2, 0, 1 },
154 { 2, 0, 1, 3 },
155 { 2, 0, 1, 3, 4 },
156 { 2, 0, 1, 4, 5, 3 },
157 };
158
159 /**
160 * Make AAC audio config object.
161 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
162 */
163 static void put_audio_specific_config(AVCodecContext *avctx)
164 {
165 PutBitContext pb;
166 AACEncContext *s = avctx->priv_data;
167
168 init_put_bits(&pb, avctx->extradata, avctx->extradata_size*8);
169 put_bits(&pb, 5, 2); //object type - AAC-LC
170 put_bits(&pb, 4, s->samplerate_index); //sample rate index
171 put_bits(&pb, 4, s->channels);
172 //GASpecificConfig
173 put_bits(&pb, 1, 0); //frame length - 1024 samples
174 put_bits(&pb, 1, 0); //does not depend on core coder
175 put_bits(&pb, 1, 0); //is not extension
176
177 //Explicitly Mark SBR absent
178 put_bits(&pb, 11, 0x2b7); //sync extension
179 put_bits(&pb, 5, AOT_SBR);
180 put_bits(&pb, 1, 0);
181 flush_put_bits(&pb);
182 }
183
184 #define WINDOW_FUNC(type) \
185 static void apply_ ##type ##_window(DSPContext *dsp, SingleChannelElement *sce, const float *audio)
186
187 WINDOW_FUNC(only_long)
188 {
189 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
190 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
191 float *out = sce->ret;
192
193 dsp->vector_fmul (out, audio, lwindow, 1024);
194 dsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
195 }
196
197 WINDOW_FUNC(long_start)
198 {
199 const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
200 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
201 float *out = sce->ret;
202
203 dsp->vector_fmul(out, audio, lwindow, 1024);
204 memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
205 dsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
206 memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
207 }
208
209 WINDOW_FUNC(long_stop)
210 {
211 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
212 const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
213 float *out = sce->ret;
214
215 memset(out, 0, sizeof(out[0]) * 448);
216 dsp->vector_fmul(out + 448, audio + 448, swindow, 128);
217 memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
218 dsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
219 }
220
221 WINDOW_FUNC(eight_short)
222 {
223 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
224 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
225 const float *in = audio + 448;
226 float *out = sce->ret;
227 int w;
228
229 for (w = 0; w < 8; w++) {
230 dsp->vector_fmul (out, in, w ? pwindow : swindow, 128);
231 out += 128;
232 in += 128;
233 dsp->vector_fmul_reverse(out, in, swindow, 128);
234 out += 128;
235 }
236 }
237
238 static void (*const apply_window[4])(DSPContext *dsp, SingleChannelElement *sce, const float *audio) = {
239 [ONLY_LONG_SEQUENCE] = apply_only_long_window,
240 [LONG_START_SEQUENCE] = apply_long_start_window,
241 [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
242 [LONG_STOP_SEQUENCE] = apply_long_stop_window
243 };
244
245 static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
246 float *audio)
247 {
248 int i;
249 float *output = sce->ret;
250
251 apply_window[sce->ics.window_sequence[0]](&s->dsp, sce, audio);
252
253 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
254 s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
255 else
256 for (i = 0; i < 1024; i += 128)
257 s->mdct128.mdct_calc(&s->mdct128, sce->coeffs + i, output + i*2);
258 memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
259 }
260
261 /**
262 * Encode ics_info element.
263 * @see Table 4.6 (syntax of ics_info)
264 */
265 static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
266 {
267 int w;
268
269 put_bits(&s->pb, 1, 0); // ics_reserved bit
270 put_bits(&s->pb, 2, info->window_sequence[0]);
271 put_bits(&s->pb, 1, info->use_kb_window[0]);
272 if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
273 put_bits(&s->pb, 6, info->max_sfb);
274 put_bits(&s->pb, 1, 0); // no prediction
275 } else {
276 put_bits(&s->pb, 4, info->max_sfb);
277 for (w = 1; w < 8; w++)
278 put_bits(&s->pb, 1, !info->group_len[w]);
279 }
280 }
281
282 /**
283 * Encode MS data.
284 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
285 */
286 static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
287 {
288 int i, w;
289
290 put_bits(pb, 2, cpe->ms_mode);
291 if (cpe->ms_mode == 1)
292 for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
293 for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
294 put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
295 }
296
297 /**
298 * Produce integer coefficients from scalefactors provided by the model.
299 */
300 static void adjust_frame_information(AACEncContext *apc, ChannelElement *cpe, int chans)
301 {
302 int i, w, w2, g, ch;
303 int start, maxsfb, cmaxsfb;
304
305 for (ch = 0; ch < chans; ch++) {
306 IndividualChannelStream *ics = &cpe->ch[ch].ics;
307 start = 0;
308 maxsfb = 0;
309 cpe->ch[ch].pulse.num_pulse = 0;
310 for (w = 0; w < ics->num_windows*16; w += 16) {
311 for (g = 0; g < ics->num_swb; g++) {
312 //apply M/S
313 if (cpe->common_window && !ch && cpe->ms_mask[w + g]) {
314 for (i = 0; i < ics->swb_sizes[g]; i++) {
315 cpe->ch[0].coeffs[start+i] = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) / 2.0;
316 cpe->ch[1].coeffs[start+i] = cpe->ch[0].coeffs[start+i] - cpe->ch[1].coeffs[start+i];
317 }
318 }
319 start += ics->swb_sizes[g];
320 }
321 for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w+cmaxsfb-1]; cmaxsfb--)
322 ;
323 maxsfb = FFMAX(maxsfb, cmaxsfb);
324 }
325 ics->max_sfb = maxsfb;
326
327 //adjust zero bands for window groups
328 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
329 for (g = 0; g < ics->max_sfb; g++) {
330 i = 1;
331 for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
332 if (!cpe->ch[ch].zeroes[w2*16 + g]) {
333 i = 0;
334 break;
335 }
336 }
337 cpe->ch[ch].zeroes[w*16 + g] = i;
338 }
339 }
340 }
341
342 if (chans > 1 && cpe->common_window) {
343 IndividualChannelStream *ics0 = &cpe->ch[0].ics;
344 IndividualChannelStream *ics1 = &cpe->ch[1].ics;
345 int msc = 0;
346 ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
347 ics1->max_sfb = ics0->max_sfb;
348 for (w = 0; w < ics0->num_windows*16; w += 16)
349 for (i = 0; i < ics0->max_sfb; i++)
350 if (cpe->ms_mask[w+i])
351 msc++;
352 if (msc == 0 || ics0->max_sfb == 0)
353 cpe->ms_mode = 0;
354 else
355 cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
356 }
357 }
358
359 /**
360 * Encode scalefactor band coding type.
361 */
362 static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
363 {
364 int w;
365
366 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
367 s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
368 }
369
370 /**
371 * Encode scalefactors.
372 */
373 static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
374 SingleChannelElement *sce)
375 {
376 int off = sce->sf_idx[0], diff;
377 int i, w;
378
379 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
380 for (i = 0; i < sce->ics.max_sfb; i++) {
381 if (!sce->zeroes[w*16 + i]) {
382 diff = sce->sf_idx[w*16 + i] - off + SCALE_DIFF_ZERO;
383 if (diff < 0 || diff > 120)
384 av_log(avctx, AV_LOG_ERROR, "Scalefactor difference is too big to be coded\n");
385 off = sce->sf_idx[w*16 + i];
386 put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
387 }
388 }
389 }
390 }
391
392 /**
393 * Encode pulse data.
394 */
395 static void encode_pulses(AACEncContext *s, Pulse *pulse)
396 {
397 int i;
398
399 put_bits(&s->pb, 1, !!pulse->num_pulse);
400 if (!pulse->num_pulse)
401 return;
402
403 put_bits(&s->pb, 2, pulse->num_pulse - 1);
404 put_bits(&s->pb, 6, pulse->start);
405 for (i = 0; i < pulse->num_pulse; i++) {
406 put_bits(&s->pb, 5, pulse->pos[i]);
407 put_bits(&s->pb, 4, pulse->amp[i]);
408 }
409 }
410
411 /**
412 * Encode spectral coefficients processed by psychoacoustic model.
413 */
414 static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
415 {
416 int start, i, w, w2;
417
418 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
419 start = 0;
420 for (i = 0; i < sce->ics.max_sfb; i++) {
421 if (sce->zeroes[w*16 + i]) {
422 start += sce->ics.swb_sizes[i];
423 continue;
424 }
425 for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++)
426 s->coder->quantize_and_encode_band(s, &s->pb, sce->coeffs + start + w2*128,
427 sce->ics.swb_sizes[i],
428 sce->sf_idx[w*16 + i],
429 sce->band_type[w*16 + i],
430 s->lambda);
431 start += sce->ics.swb_sizes[i];
432 }
433 }
434 }
435
436 /**
437 * Encode one channel of audio data.
438 */
439 static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
440 SingleChannelElement *sce,
441 int common_window)
442 {
443 put_bits(&s->pb, 8, sce->sf_idx[0]);
444 if (!common_window)
445 put_ics_info(s, &sce->ics);
446 encode_band_info(s, sce);
447 encode_scale_factors(avctx, s, sce);
448 encode_pulses(s, &sce->pulse);
449 put_bits(&s->pb, 1, 0); //tns
450 put_bits(&s->pb, 1, 0); //ssr
451 encode_spectral_coeffs(s, sce);
452 return 0;
453 }
454
455 /**
456 * Write some auxiliary information about the created AAC file.
457 */
458 static void put_bitstream_info(AVCodecContext *avctx, AACEncContext *s,
459 const char *name)
460 {
461 int i, namelen, padbits;
462
463 namelen = strlen(name) + 2;
464 put_bits(&s->pb, 3, TYPE_FIL);
465 put_bits(&s->pb, 4, FFMIN(namelen, 15));
466 if (namelen >= 15)
467 put_bits(&s->pb, 8, namelen - 14);
468 put_bits(&s->pb, 4, 0); //extension type - filler
469 padbits = -put_bits_count(&s->pb) & 7;
470 avpriv_align_put_bits(&s->pb);
471 for (i = 0; i < namelen - 2; i++)
472 put_bits(&s->pb, 8, name[i]);
473 put_bits(&s->pb, 12 - padbits, 0);
474 }
475
476 /*
477 * Deinterleave input samples.
478 * Channels are reordered from Libav's default order to AAC order.
479 */
480 static void deinterleave_input_samples(AACEncContext *s, const AVFrame *frame)
481 {
482 int ch, i;
483 const int sinc = s->channels;
484 const uint8_t *channel_map = aac_chan_maps[sinc - 1];
485
486 /* deinterleave and remap input samples */
487 for (ch = 0; ch < sinc; ch++) {
488 /* copy last 1024 samples of previous frame to the start of the current frame */
489 memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
490
491 /* deinterleave */
492 i = 2048;
493 if (frame) {
494 const float *sptr = ((const float *)frame->data[0]) + channel_map[ch];
495 for (; i < 2048 + frame->nb_samples; i++) {
496 s->planar_samples[ch][i] = *sptr;
497 sptr += sinc;
498 }
499 }
500 memset(&s->planar_samples[ch][i], 0,
501 (3072 - i) * sizeof(s->planar_samples[0][0]));
502 }
503 }
504
505 static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
506 const AVFrame *frame, int *got_packet_ptr)
507 {
508 AACEncContext *s = avctx->priv_data;
509 float **samples = s->planar_samples, *samples2, *la, *overlap;
510 ChannelElement *cpe;
511 int i, ch, w, g, chans, tag, start_ch, ret;
512 int chan_el_counter[4];
513 FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
514
515 if (s->last_frame == 2)
516 return 0;
517
518 /* add current frame to queue */
519 if (frame) {
520 if ((ret = ff_af_queue_add(&s->afq, frame) < 0))
521 return ret;
522 }
523
524 deinterleave_input_samples(s, frame);
525 if (s->psypp)
526 ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
527
528 if (!avctx->frame_number)
529 return 0;
530
531 start_ch = 0;
532 for (i = 0; i < s->chan_map[0]; i++) {
533 FFPsyWindowInfo* wi = windows + start_ch;
534 tag = s->chan_map[i+1];
535 chans = tag == TYPE_CPE ? 2 : 1;
536 cpe = &s->cpe[i];
537 for (ch = 0; ch < chans; ch++) {
538 IndividualChannelStream *ics = &cpe->ch[ch].ics;
539 int cur_channel = start_ch + ch;
540 overlap = &samples[cur_channel][0];
541 samples2 = overlap + 1024;
542 la = samples2 + (448+64);
543 if (!frame)
544 la = NULL;
545 if (tag == TYPE_LFE) {
546 wi[ch].window_type[0] = ONLY_LONG_SEQUENCE;
547 wi[ch].window_shape = 0;
548 wi[ch].num_windows = 1;
549 wi[ch].grouping[0] = 1;
550
551 /* Only the lowest 12 coefficients are used in a LFE channel.
552 * The expression below results in only the bottom 8 coefficients
553 * being used for 11.025kHz to 16kHz sample rates.
554 */
555 ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
556 } else {
557 wi[ch] = s->psy.model->window(&s->psy, samples2, la, cur_channel,
558 ics->window_sequence[0]);
559 }
560 ics->window_sequence[1] = ics->window_sequence[0];
561 ics->window_sequence[0] = wi[ch].window_type[0];
562 ics->use_kb_window[1] = ics->use_kb_window[0];
563 ics->use_kb_window[0] = wi[ch].window_shape;
564 ics->num_windows = wi[ch].num_windows;
565 ics->swb_sizes = s->psy.bands [ics->num_windows == 8];
566 ics->num_swb = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
567 for (w = 0; w < ics->num_windows; w++)
568 ics->group_len[w] = wi[ch].grouping[w];
569
570 apply_window_and_mdct(s, &cpe->ch[ch], overlap);
571 }
572 start_ch += chans;
573 }
574 do {
575 int frame_bits;
576
577 if ((ret = ff_alloc_packet(avpkt, 768 * s->channels))) {
578 av_log(avctx, AV_LOG_ERROR, "Error getting output packet\n");
579 return ret;
580 }
581 init_put_bits(&s->pb, avpkt->data, avpkt->size);
582
583 if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & CODEC_FLAG_BITEXACT))
584 put_bitstream_info(avctx, s, LIBAVCODEC_IDENT);
585 start_ch = 0;
586 memset(chan_el_counter, 0, sizeof(chan_el_counter));
587 for (i = 0; i < s->chan_map[0]; i++) {
588 FFPsyWindowInfo* wi = windows + start_ch;
589 const float *coeffs[2];
590 tag = s->chan_map[i+1];
591 chans = tag == TYPE_CPE ? 2 : 1;
592 cpe = &s->cpe[i];
593 put_bits(&s->pb, 3, tag);
594 put_bits(&s->pb, 4, chan_el_counter[tag]++);
595 for (ch = 0; ch < chans; ch++)
596 coeffs[ch] = cpe->ch[ch].coeffs;
597 s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
598 for (ch = 0; ch < chans; ch++) {
599 s->cur_channel = start_ch * 2 + ch;
600 s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
601 }
602 cpe->common_window = 0;
603 if (chans > 1
604 && wi[0].window_type[0] == wi[1].window_type[0]
605 && wi[0].window_shape == wi[1].window_shape) {
606
607 cpe->common_window = 1;
608 for (w = 0; w < wi[0].num_windows; w++) {
609 if (wi[0].grouping[w] != wi[1].grouping[w]) {
610 cpe->common_window = 0;
611 break;
612 }
613 }
614 }
615 s->cur_channel = start_ch * 2;
616 if (s->options.stereo_mode && cpe->common_window) {
617 if (s->options.stereo_mode > 0) {
618 IndividualChannelStream *ics = &cpe->ch[0].ics;
619 for (w = 0; w < ics->num_windows; w += ics->group_len[w])
620 for (g = 0; g < ics->num_swb; g++)
621 cpe->ms_mask[w*16+g] = 1;
622 } else if (s->coder->search_for_ms) {
623 s->coder->search_for_ms(s, cpe, s->lambda);
624 }
625 }
626 adjust_frame_information(s, cpe, chans);
627 if (chans == 2) {
628 put_bits(&s->pb, 1, cpe->common_window);
629 if (cpe->common_window) {
630 put_ics_info(s, &cpe->ch[0].ics);
631 encode_ms_info(&s->pb, cpe);
632 }
633 }
634 for (ch = 0; ch < chans; ch++) {
635 s->cur_channel = start_ch + ch;
636 encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
637 }
638 start_ch += chans;
639 }
640
641 frame_bits = put_bits_count(&s->pb);
642 if (frame_bits <= 6144 * s->channels - 3) {
643 s->psy.bitres.bits = frame_bits / s->channels;
644 break;
645 }
646
647 s->lambda *= avctx->bit_rate * 1024.0f / avctx->sample_rate / frame_bits;
648
649 } while (1);
650
651 put_bits(&s->pb, 3, TYPE_END);
652 flush_put_bits(&s->pb);
653 avctx->frame_bits = put_bits_count(&s->pb);
654
655 // rate control stuff
656 if (!(avctx->flags & CODEC_FLAG_QSCALE)) {
657 float ratio = avctx->bit_rate * 1024.0f / avctx->sample_rate / avctx->frame_bits;
658 s->lambda *= ratio;
659 s->lambda = FFMIN(s->lambda, 65536.f);
660 }
661
662 if (!frame)
663 s->last_frame++;
664
665 ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
666 &avpkt->duration);
667
668 avpkt->size = put_bits_count(&s->pb) >> 3;
669 *got_packet_ptr = 1;
670 return 0;
671 }
672
673 static av_cold int aac_encode_end(AVCodecContext *avctx)
674 {
675 AACEncContext *s = avctx->priv_data;
676
677 ff_mdct_end(&s->mdct1024);
678 ff_mdct_end(&s->mdct128);
679 ff_psy_end(&s->psy);
680 if (s->psypp)
681 ff_psy_preprocess_end(s->psypp);
682 av_freep(&s->buffer.samples);
683 av_freep(&s->cpe);
684 ff_af_queue_close(&s->afq);
685 #if FF_API_OLD_ENCODE_AUDIO
686 av_freep(&avctx->coded_frame);
687 #endif
688 return 0;
689 }
690
691 static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
692 {
693 int ret = 0;
694
695 ff_dsputil_init(&s->dsp, avctx);
696
697 // window init
698 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
699 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
700 ff_init_ff_sine_windows(10);
701 ff_init_ff_sine_windows(7);
702
703 if (ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0))
704 return ret;
705 if (ret = ff_mdct_init(&s->mdct128, 8, 0, 32768.0))
706 return ret;
707
708 return 0;
709 }
710
711 static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
712 {
713 int ch;
714 FF_ALLOCZ_OR_GOTO(avctx, s->buffer.samples, 3 * 1024 * s->channels * sizeof(s->buffer.samples[0]), alloc_fail);
715 FF_ALLOCZ_OR_GOTO(avctx, s->cpe, sizeof(ChannelElement) * s->chan_map[0], alloc_fail);
716 FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + FF_INPUT_BUFFER_PADDING_SIZE, alloc_fail);
717
718 for(ch = 0; ch < s->channels; ch++)
719 s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
720
721 #if FF_API_OLD_ENCODE_AUDIO
722 if (!(avctx->coded_frame = avcodec_alloc_frame()))
723 goto alloc_fail;
724 #endif
725
726 return 0;
727 alloc_fail:
728 return AVERROR(ENOMEM);
729 }
730
731 static av_cold int aac_encode_init(AVCodecContext *avctx)
732 {
733 AACEncContext *s = avctx->priv_data;
734 int i, ret = 0;
735 const uint8_t *sizes[2];
736 uint8_t grouping[AAC_MAX_CHANNELS];
737 int lengths[2];
738
739 avctx->frame_size = 1024;
740
741 for (i = 0; i < 16; i++)
742 if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
743 break;
744
745 s->channels = avctx->channels;
746
747 ERROR_IF(i == 16,
748 "Unsupported sample rate %d\n", avctx->sample_rate);
749 ERROR_IF(s->channels > AAC_MAX_CHANNELS,
750 "Unsupported number of channels: %d\n", s->channels);
751 ERROR_IF(avctx->profile != FF_PROFILE_UNKNOWN && avctx->profile != FF_PROFILE_AAC_LOW,
752 "Unsupported profile %d\n", avctx->profile);
753 ERROR_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
754 "Too many bits per frame requested\n");
755
756 s->samplerate_index = i;
757
758 s->chan_map = aac_chan_configs[s->channels-1];
759
760 if (ret = dsp_init(avctx, s))
761 goto fail;
762
763 if (ret = alloc_buffers(avctx, s))
764 goto fail;
765
766 avctx->extradata_size = 5;
767 put_audio_specific_config(avctx);
768
769 sizes[0] = swb_size_1024[i];
770 sizes[1] = swb_size_128[i];
771 lengths[0] = ff_aac_num_swb_1024[i];
772 lengths[1] = ff_aac_num_swb_128[i];
773 for (i = 0; i < s->chan_map[0]; i++)
774 grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
775 if (ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths, s->chan_map[0], grouping))
776 goto fail;
777 s->psypp = ff_psy_preprocess_init(avctx);
778 s->coder = &ff_aac_coders[2];
779
780 s->lambda = avctx->global_quality ? avctx->global_quality : 120;
781
782 ff_aac_tableinit();
783
784 for (i = 0; i < 428; i++)
785 ff_aac_pow34sf_tab[i] = sqrt(ff_aac_pow2sf_tab[i] * sqrt(ff_aac_pow2sf_tab[i]));
786
787 avctx->delay = 1024;
788 ff_af_queue_init(avctx, &s->afq);
789
790 return 0;
791 fail:
792 aac_encode_end(avctx);
793 return ret;
794 }
795
796 #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
797 static const AVOption aacenc_options[] = {
798 {"stereo_mode", "Stereo coding method", offsetof(AACEncContext, options.stereo_mode), AV_OPT_TYPE_INT, {.dbl = 0}, -1, 1, AACENC_FLAGS, "stereo_mode"},
799 {"auto", "Selected by the Encoder", 0, AV_OPT_TYPE_CONST, {.dbl = -1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
800 {"ms_off", "Disable Mid/Side coding", 0, AV_OPT_TYPE_CONST, {.dbl = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
801 {"ms_force", "Force Mid/Side for the whole frame if possible", 0, AV_OPT_TYPE_CONST, {.dbl = 1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
802 {NULL}
803 };
804
805 static const AVClass aacenc_class = {
806 "AAC encoder",
807 av_default_item_name,
808 aacenc_options,
809 LIBAVUTIL_VERSION_INT,
810 };
811
812 AVCodec ff_aac_encoder = {
813 .name = "aac",
814 .type = AVMEDIA_TYPE_AUDIO,
815 .id = CODEC_ID_AAC,
816 .priv_data_size = sizeof(AACEncContext),
817 .init = aac_encode_init,
818 .encode2 = aac_encode_frame,
819 .close = aac_encode_end,
820 .capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY | CODEC_CAP_EXPERIMENTAL,
821 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_FLT,AV_SAMPLE_FMT_NONE},
822 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
823 .priv_class = &aacenc_class,
824 };