FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
aacenc.c
Go to the documentation of this file.
1 /*
2  * AAC encoder
3  * Copyright (C) 2008 Konstantin Shishkov
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  * AAC encoder
25  */
26 
27 /***********************************
28  * TODOs:
29  * add sane pulse detection
30  * add temporal noise shaping
31  ***********************************/
32 
33 #include "libavutil/float_dsp.h"
34 #include "libavutil/opt.h"
35 #include "avcodec.h"
36 #include "put_bits.h"
37 #include "dsputil.h"
38 #include "internal.h"
39 #include "mpeg4audio.h"
40 #include "kbdwin.h"
41 #include "sinewin.h"
42 
43 #include "aac.h"
44 #include "aactab.h"
45 #include "aacenc.h"
46 
47 #include "psymodel.h"
48 
49 #define AAC_MAX_CHANNELS 6
50 
51 #define ERROR_IF(cond, ...) \
52  if (cond) { \
53  av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \
54  return AVERROR(EINVAL); \
55  }
56 
57 float ff_aac_pow34sf_tab[428];
58 
59 static const uint8_t swb_size_1024_96[] = {
60  4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8,
61  12, 12, 12, 12, 12, 16, 16, 24, 28, 36, 44,
62  64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64
63 };
64 
65 static const uint8_t swb_size_1024_64[] = {
66  4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8,
67  12, 12, 12, 16, 16, 16, 20, 24, 24, 28, 36,
68  40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40
69 };
70 
71 static const uint8_t swb_size_1024_48[] = {
72  4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8,
73  12, 12, 12, 12, 16, 16, 20, 20, 24, 24, 28, 28,
74  32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
75  96
76 };
77 
78 static const uint8_t swb_size_1024_32[] = {
79  4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8,
80  12, 12, 12, 12, 16, 16, 20, 20, 24, 24, 28, 28,
81  32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32
82 };
83 
84 static const uint8_t swb_size_1024_24[] = {
85  4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
86  12, 12, 12, 12, 16, 16, 16, 20, 20, 24, 24, 28, 28,
87  32, 36, 36, 40, 44, 48, 52, 52, 64, 64, 64, 64, 64
88 };
89 
90 static const uint8_t swb_size_1024_16[] = {
91  8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
92  12, 12, 12, 12, 12, 12, 12, 12, 12, 16, 16, 16, 16, 20, 20, 20, 24, 24, 28, 28,
93  32, 36, 40, 40, 44, 48, 52, 56, 60, 64, 64, 64
94 };
95 
96 static const uint8_t swb_size_1024_8[] = {
97  12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12,
98  16, 16, 16, 16, 16, 16, 16, 20, 20, 20, 20, 24, 24, 24, 28, 28,
99  32, 36, 36, 40, 44, 48, 52, 56, 60, 64, 80
100 };
101 
102 static const uint8_t *swb_size_1024[] = {
103  swb_size_1024_96, swb_size_1024_96, swb_size_1024_64,
104  swb_size_1024_48, swb_size_1024_48, swb_size_1024_32,
105  swb_size_1024_24, swb_size_1024_24, swb_size_1024_16,
106  swb_size_1024_16, swb_size_1024_16, swb_size_1024_8
107 };
108 
109 static const uint8_t swb_size_128_96[] = {
110  4, 4, 4, 4, 4, 4, 8, 8, 8, 16, 28, 36
111 };
112 
113 static const uint8_t swb_size_128_48[] = {
114  4, 4, 4, 4, 4, 8, 8, 8, 12, 12, 12, 16, 16, 16
115 };
116 
117 static const uint8_t swb_size_128_24[] = {
118  4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 20
119 };
120 
121 static const uint8_t swb_size_128_16[] = {
122  4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 12, 12, 16, 20, 20
123 };
124 
125 static const uint8_t swb_size_128_8[] = {
126  4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 12, 16, 20, 20
127 };
128 
129 static const uint8_t *swb_size_128[] = {
130  /* the last entry on the following row is swb_size_128_64 but is a
131  duplicate of swb_size_128_96 */
132  swb_size_128_96, swb_size_128_96, swb_size_128_96,
133  swb_size_128_48, swb_size_128_48, swb_size_128_48,
134  swb_size_128_24, swb_size_128_24, swb_size_128_16,
135  swb_size_128_16, swb_size_128_16, swb_size_128_8
136 };
137 
138 /** default channel configurations */
139 static const uint8_t aac_chan_configs[6][5] = {
140  {1, TYPE_SCE}, // 1 channel - single channel element
141  {1, TYPE_CPE}, // 2 channels - channel pair
142  {2, TYPE_SCE, TYPE_CPE}, // 3 channels - center + stereo
143  {3, TYPE_SCE, TYPE_CPE, TYPE_SCE}, // 4 channels - front center + stereo + back center
144  {3, TYPE_SCE, TYPE_CPE, TYPE_CPE}, // 5 channels - front center + stereo + back stereo
145  {4, TYPE_SCE, TYPE_CPE, TYPE_CPE, TYPE_LFE}, // 6 channels - front center + stereo + back stereo + LFE
146 };
147 
148 /**
149  * Table to remap channels from libavcodec's default order to AAC order.
150  */
151 static const uint8_t aac_chan_maps[AAC_MAX_CHANNELS][AAC_MAX_CHANNELS] = {
152  { 0 },
153  { 0, 1 },
154  { 2, 0, 1 },
155  { 2, 0, 1, 3 },
156  { 2, 0, 1, 3, 4 },
157  { 2, 0, 1, 4, 5, 3 },
158 };
159 
160 /**
161  * Make AAC audio config object.
162  * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
163  */
164 static void put_audio_specific_config(AVCodecContext *avctx)
165 {
166  PutBitContext pb;
167  AACEncContext *s = avctx->priv_data;
168 
169  init_put_bits(&pb, avctx->extradata, avctx->extradata_size*8);
170  put_bits(&pb, 5, 2); //object type - AAC-LC
171  put_bits(&pb, 4, s->samplerate_index); //sample rate index
172  put_bits(&pb, 4, s->channels);
173  //GASpecificConfig
174  put_bits(&pb, 1, 0); //frame length - 1024 samples
175  put_bits(&pb, 1, 0); //does not depend on core coder
176  put_bits(&pb, 1, 0); //is not extension
177 
178  //Explicitly Mark SBR absent
179  put_bits(&pb, 11, 0x2b7); //sync extension
180  put_bits(&pb, 5, AOT_SBR);
181  put_bits(&pb, 1, 0);
182  flush_put_bits(&pb);
183 }
184 
185 #define WINDOW_FUNC(type) \
186 static void apply_ ##type ##_window(DSPContext *dsp, AVFloatDSPContext *fdsp, \
187  SingleChannelElement *sce, \
188  const float *audio)
189 
190 WINDOW_FUNC(only_long)
191 {
192  const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
193  const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
194  float *out = sce->ret_buf;
195 
196  fdsp->vector_fmul (out, audio, lwindow, 1024);
197  dsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
198 }
199 
200 WINDOW_FUNC(long_start)
201 {
202  const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
203  const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
204  float *out = sce->ret_buf;
205 
206  fdsp->vector_fmul(out, audio, lwindow, 1024);
207  memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
208  dsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
209  memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
210 }
211 
212 WINDOW_FUNC(long_stop)
213 {
214  const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
215  const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
216  float *out = sce->ret_buf;
217 
218  memset(out, 0, sizeof(out[0]) * 448);
219  fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
220  memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
221  dsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
222 }
223 
224 WINDOW_FUNC(eight_short)
225 {
226  const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
227  const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
228  const float *in = audio + 448;
229  float *out = sce->ret_buf;
230  int w;
231 
232  for (w = 0; w < 8; w++) {
233  fdsp->vector_fmul (out, in, w ? pwindow : swindow, 128);
234  out += 128;
235  in += 128;
236  dsp->vector_fmul_reverse(out, in, swindow, 128);
237  out += 128;
238  }
239 }
240 
241 static void (*const apply_window[4])(DSPContext *dsp, AVFloatDSPContext *fdsp,
242  SingleChannelElement *sce,
243  const float *audio) = {
244  [ONLY_LONG_SEQUENCE] = apply_only_long_window,
245  [LONG_START_SEQUENCE] = apply_long_start_window,
246  [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
247  [LONG_STOP_SEQUENCE] = apply_long_stop_window
248 };
249 
250 static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
251  float *audio)
252 {
253  int i;
254  float *output = sce->ret_buf;
255 
256  apply_window[sce->ics.window_sequence[0]](&s->dsp, &s->fdsp, sce, audio);
257 
258  if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
259  s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
260  else
261  for (i = 0; i < 1024; i += 128)
262  s->mdct128.mdct_calc(&s->mdct128, sce->coeffs + i, output + i*2);
263  memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
264 }
265 
266 /**
267  * Encode ics_info element.
268  * @see Table 4.6 (syntax of ics_info)
269  */
270 static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
271 {
272  int w;
273 
274  put_bits(&s->pb, 1, 0); // ics_reserved bit
275  put_bits(&s->pb, 2, info->window_sequence[0]);
276  put_bits(&s->pb, 1, info->use_kb_window[0]);
277  if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
278  put_bits(&s->pb, 6, info->max_sfb);
279  put_bits(&s->pb, 1, 0); // no prediction
280  } else {
281  put_bits(&s->pb, 4, info->max_sfb);
282  for (w = 1; w < 8; w++)
283  put_bits(&s->pb, 1, !info->group_len[w]);
284  }
285 }
286 
287 /**
288  * Encode MS data.
289  * @see 4.6.8.1 "Joint Coding - M/S Stereo"
290  */
291 static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
292 {
293  int i, w;
294 
295  put_bits(pb, 2, cpe->ms_mode);
296  if (cpe->ms_mode == 1)
297  for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
298  for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
299  put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
300 }
301 
302 /**
303  * Produce integer coefficients from scalefactors provided by the model.
304  */
305 static void adjust_frame_information(ChannelElement *cpe, int chans)
306 {
307  int i, w, w2, g, ch;
308  int start, maxsfb, cmaxsfb;
309 
310  for (ch = 0; ch < chans; ch++) {
311  IndividualChannelStream *ics = &cpe->ch[ch].ics;
312  start = 0;
313  maxsfb = 0;
314  cpe->ch[ch].pulse.num_pulse = 0;
315  for (w = 0; w < ics->num_windows*16; w += 16) {
316  for (g = 0; g < ics->num_swb; g++) {
317  //apply M/S
318  if (cpe->common_window && !ch && cpe->ms_mask[w + g]) {
319  for (i = 0; i < ics->swb_sizes[g]; i++) {
320  cpe->ch[0].coeffs[start+i] = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) / 2.0;
321  cpe->ch[1].coeffs[start+i] = cpe->ch[0].coeffs[start+i] - cpe->ch[1].coeffs[start+i];
322  }
323  }
324  start += ics->swb_sizes[g];
325  }
326  for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w+cmaxsfb-1]; cmaxsfb--)
327  ;
328  maxsfb = FFMAX(maxsfb, cmaxsfb);
329  }
330  ics->max_sfb = maxsfb;
331 
332  //adjust zero bands for window groups
333  for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
334  for (g = 0; g < ics->max_sfb; g++) {
335  i = 1;
336  for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
337  if (!cpe->ch[ch].zeroes[w2*16 + g]) {
338  i = 0;
339  break;
340  }
341  }
342  cpe->ch[ch].zeroes[w*16 + g] = i;
343  }
344  }
345  }
346 
347  if (chans > 1 && cpe->common_window) {
348  IndividualChannelStream *ics0 = &cpe->ch[0].ics;
349  IndividualChannelStream *ics1 = &cpe->ch[1].ics;
350  int msc = 0;
351  ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
352  ics1->max_sfb = ics0->max_sfb;
353  for (w = 0; w < ics0->num_windows*16; w += 16)
354  for (i = 0; i < ics0->max_sfb; i++)
355  if (cpe->ms_mask[w+i])
356  msc++;
357  if (msc == 0 || ics0->max_sfb == 0)
358  cpe->ms_mode = 0;
359  else
360  cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
361  }
362 }
363 
364 /**
365  * Encode scalefactor band coding type.
366  */
367 static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
368 {
369  int w;
370 
371  for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
372  s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
373 }
374 
375 /**
376  * Encode scalefactors.
377  */
378 static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
379  SingleChannelElement *sce)
380 {
381  int off = sce->sf_idx[0], diff;
382  int i, w;
383 
384  for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
385  for (i = 0; i < sce->ics.max_sfb; i++) {
386  if (!sce->zeroes[w*16 + i]) {
387  diff = sce->sf_idx[w*16 + i] - off + SCALE_DIFF_ZERO;
388  av_assert0(diff >= 0 && diff <= 120);
389  off = sce->sf_idx[w*16 + i];
390  put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
391  }
392  }
393  }
394 }
395 
396 /**
397  * Encode pulse data.
398  */
399 static void encode_pulses(AACEncContext *s, Pulse *pulse)
400 {
401  int i;
402 
403  put_bits(&s->pb, 1, !!pulse->num_pulse);
404  if (!pulse->num_pulse)
405  return;
406 
407  put_bits(&s->pb, 2, pulse->num_pulse - 1);
408  put_bits(&s->pb, 6, pulse->start);
409  for (i = 0; i < pulse->num_pulse; i++) {
410  put_bits(&s->pb, 5, pulse->pos[i]);
411  put_bits(&s->pb, 4, pulse->amp[i]);
412  }
413 }
414 
415 /**
416  * Encode spectral coefficients processed by psychoacoustic model.
417  */
418 static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
419 {
420  int start, i, w, w2;
421 
422  for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
423  start = 0;
424  for (i = 0; i < sce->ics.max_sfb; i++) {
425  if (sce->zeroes[w*16 + i]) {
426  start += sce->ics.swb_sizes[i];
427  continue;
428  }
429  for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++)
430  s->coder->quantize_and_encode_band(s, &s->pb, sce->coeffs + start + w2*128,
431  sce->ics.swb_sizes[i],
432  sce->sf_idx[w*16 + i],
433  sce->band_type[w*16 + i],
434  s->lambda);
435  start += sce->ics.swb_sizes[i];
436  }
437  }
438 }
439 
440 /**
441  * Encode one channel of audio data.
442  */
443 static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
444  SingleChannelElement *sce,
445  int common_window)
446 {
447  put_bits(&s->pb, 8, sce->sf_idx[0]);
448  if (!common_window)
449  put_ics_info(s, &sce->ics);
450  encode_band_info(s, sce);
451  encode_scale_factors(avctx, s, sce);
452  encode_pulses(s, &sce->pulse);
453  put_bits(&s->pb, 1, 0); //tns
454  put_bits(&s->pb, 1, 0); //ssr
455  encode_spectral_coeffs(s, sce);
456  return 0;
457 }
458 
459 /**
460  * Write some auxiliary information about the created AAC file.
461  */
462 static void put_bitstream_info(AACEncContext *s, const char *name)
463 {
464  int i, namelen, padbits;
465 
466  namelen = strlen(name) + 2;
467  put_bits(&s->pb, 3, TYPE_FIL);
468  put_bits(&s->pb, 4, FFMIN(namelen, 15));
469  if (namelen >= 15)
470  put_bits(&s->pb, 8, namelen - 14);
471  put_bits(&s->pb, 4, 0); //extension type - filler
472  padbits = -put_bits_count(&s->pb) & 7;
473  avpriv_align_put_bits(&s->pb);
474  for (i = 0; i < namelen - 2; i++)
475  put_bits(&s->pb, 8, name[i]);
476  put_bits(&s->pb, 12 - padbits, 0);
477 }
478 
479 /*
480  * Copy input samples.
481  * Channels are reordered from libavcodec's default order to AAC order.
482  */
483 static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
484 {
485  int ch;
486  int end = 2048 + (frame ? frame->nb_samples : 0);
487  const uint8_t *channel_map = aac_chan_maps[s->channels - 1];
488 
489  /* copy and remap input samples */
490  for (ch = 0; ch < s->channels; ch++) {
491  /* copy last 1024 samples of previous frame to the start of the current frame */
492  memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
493 
494  /* copy new samples and zero any remaining samples */
495  if (frame) {
496  memcpy(&s->planar_samples[ch][2048],
497  frame->extended_data[channel_map[ch]],
498  frame->nb_samples * sizeof(s->planar_samples[0][0]));
499  }
500  memset(&s->planar_samples[ch][end], 0,
501  (3072 - end) * 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  copy_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  if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels))) {
575  av_log(avctx, AV_LOG_ERROR, "Error getting output packet\n");
576  return ret;
577  }
578  do {
579  int frame_bits;
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(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(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  avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
697 
698  // window init
699  ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
700  ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
701  ff_init_ff_sine_windows(10);
702  ff_init_ff_sine_windows(7);
703 
704  if (ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0))
705  return ret;
706  if (ret = ff_mdct_init(&s->mdct128, 8, 0, 32768.0))
707  return ret;
708 
709  return 0;
710 }
711 
712 static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
713 {
714  int ch;
715  FF_ALLOCZ_OR_GOTO(avctx, s->buffer.samples, 3 * 1024 * s->channels * sizeof(s->buffer.samples[0]), alloc_fail);
716  FF_ALLOCZ_OR_GOTO(avctx, s->cpe, sizeof(ChannelElement) * s->chan_map[0], alloc_fail);
717  FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + FF_INPUT_BUFFER_PADDING_SIZE, alloc_fail);
718 
719  for(ch = 0; ch < s->channels; ch++)
720  s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
721 
722 #if FF_API_OLD_ENCODE_AUDIO
723  if (!(avctx->coded_frame = avcodec_alloc_frame()))
724  goto alloc_fail;
725 #endif
726 
727  return 0;
728 alloc_fail:
729  return AVERROR(ENOMEM);
730 }
731 
732 static av_cold int aac_encode_init(AVCodecContext *avctx)
733 {
734  AACEncContext *s = avctx->priv_data;
735  int i, ret = 0;
736  const uint8_t *sizes[2];
737  uint8_t grouping[AAC_MAX_CHANNELS];
738  int lengths[2];
739 
740  avctx->frame_size = 1024;
741 
742  for (i = 0; i < 16; i++)
743  if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
744  break;
745 
746  s->channels = avctx->channels;
747 
748  ERROR_IF(i == 16,
749  "Unsupported sample rate %d\n", avctx->sample_rate);
750  ERROR_IF(s->channels > AAC_MAX_CHANNELS,
751  "Unsupported number of channels: %d\n", s->channels);
752  ERROR_IF(avctx->profile != FF_PROFILE_UNKNOWN && avctx->profile != FF_PROFILE_AAC_LOW,
753  "Unsupported profile %d\n", avctx->profile);
754  ERROR_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
755  "Too many bits per frame requested\n");
756 
757  s->samplerate_index = i;
758 
759  s->chan_map = aac_chan_configs[s->channels-1];
760 
761  if (ret = dsp_init(avctx, s))
762  goto fail;
763 
764  if (ret = alloc_buffers(avctx, s))
765  goto fail;
766 
767  avctx->extradata_size = 5;
768  put_audio_specific_config(avctx);
769 
770  sizes[0] = swb_size_1024[i];
771  sizes[1] = swb_size_128[i];
772  lengths[0] = ff_aac_num_swb_1024[i];
773  lengths[1] = ff_aac_num_swb_128[i];
774  for (i = 0; i < s->chan_map[0]; i++)
775  grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
776  if (ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths, s->chan_map[0], grouping))
777  goto fail;
778  s->psypp = ff_psy_preprocess_init(avctx);
779  s->coder = &ff_aac_coders[s->options.aac_coder];
780 
781  s->lambda = avctx->global_quality ? avctx->global_quality : 120;
782 
783  ff_aac_tableinit();
784 
785  for (i = 0; i < 428; i++)
786  ff_aac_pow34sf_tab[i] = sqrt(ff_aac_pow2sf_tab[i] * sqrt(ff_aac_pow2sf_tab[i]));
787 
788  avctx->delay = 1024;
789  ff_af_queue_init(avctx, &s->afq);
790 
791  return 0;
792 fail:
793  aac_encode_end(avctx);
794  return ret;
795 }
796 
797 #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
798 static const AVOption aacenc_options[] = {
799  {"stereo_mode", "Stereo coding method", offsetof(AACEncContext, options.stereo_mode), AV_OPT_TYPE_INT, {.i64 = 0}, -1, 1, AACENC_FLAGS, "stereo_mode"},
800  {"auto", "Selected by the Encoder", 0, AV_OPT_TYPE_CONST, {.i64 = -1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
801  {"ms_off", "Disable Mid/Side coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
802  {"ms_force", "Force Mid/Side for the whole frame if possible", 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
803  {"aac_coder", "", offsetof(AACEncContext, options.aac_coder), AV_OPT_TYPE_INT, {.i64 = 2}, 0, AAC_CODER_NB-1, AACENC_FLAGS},
804  {NULL}
805 };
806 
807 static const AVClass aacenc_class = {
808  "AAC encoder",
809  av_default_item_name,
810  aacenc_options,
811  LIBAVUTIL_VERSION_INT,
812 };
813 
814 /* duplicated from avpriv_mpeg4audio_sample_rates to avoid shared build
815  * failures */
816 static const int mpeg4audio_sample_rates[16] = {
817  96000, 88200, 64000, 48000, 44100, 32000,
818  24000, 22050, 16000, 12000, 11025, 8000, 7350
819 };
820 
821 AVCodec ff_aac_encoder = {
822  .name = "aac",
823  .type = AVMEDIA_TYPE_AUDIO,
824  .id = AV_CODEC_ID_AAC,
825  .priv_data_size = sizeof(AACEncContext),
826  .init = aac_encode_init,
827  .encode2 = aac_encode_frame,
828  .close = aac_encode_end,
829  .supported_samplerates = mpeg4audio_sample_rates,
830  .capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY |
831  CODEC_CAP_EXPERIMENTAL,
832  .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
833  AV_SAMPLE_FMT_NONE },
834  .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
835  .priv_class = &aacenc_class,
836 };
Generated on Sat May 25 2013 03:58:30 for FFmpeg by   doxygen 1.8.2