3 * Copyright (C) 2008 Konstantin Shishkov
5 * This file is part of FFmpeg.
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.
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.
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
27 /***********************************
29 * add sane pulse detection
30 * add temporal noise shaping
31 ***********************************/
33 #include "libavutil/float_dsp.h"
34 #include "libavutil/opt.h"
38 #include "mpeg4audio.h"
48 #define AAC_MAX_CHANNELS 6
50 #define ERROR_IF(cond, ...) \
52 av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \
53 return AVERROR(EINVAL); \
56 float ff_aac_pow34sf_tab[428];
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
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
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,
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
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
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
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
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
108 static const uint8_t swb_size_128_96[] = {
109 4, 4, 4, 4, 4, 4, 8, 8, 8, 16, 28, 36
112 static const uint8_t swb_size_128_48[] = {
113 4, 4, 4, 4, 4, 8, 8, 8, 12, 12, 12, 16, 16, 16
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
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
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
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
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
148 * Table to remap channels from libavcodec's default order to AAC order.
150 static const uint8_t aac_chan_maps[AAC_MAX_CHANNELS][AAC_MAX_CHANNELS] = {
156 { 2, 0, 1, 4, 5, 3 },
160 * Make AAC audio config object.
161 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
163 static void put_audio_specific_config(AVCodecContext *avctx)
166 AACEncContext *s = avctx->priv_data;
168 init_put_bits(&pb, avctx->extradata, avctx->extradata_size);
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);
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
177 //Explicitly Mark SBR absent
178 put_bits(&pb, 11, 0x2b7); //sync extension
179 put_bits(&pb, 5, AOT_SBR);
184 #define WINDOW_FUNC(type) \
185 static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \
186 SingleChannelElement *sce, \
189 WINDOW_FUNC(only_long)
191 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
192 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
193 float *out = sce->ret_buf;
195 fdsp->vector_fmul (out, audio, lwindow, 1024);
196 fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
199 WINDOW_FUNC(long_start)
201 const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
202 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
203 float *out = sce->ret_buf;
205 fdsp->vector_fmul(out, audio, lwindow, 1024);
206 memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
207 fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
208 memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
211 WINDOW_FUNC(long_stop)
213 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
214 const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
215 float *out = sce->ret_buf;
217 memset(out, 0, sizeof(out[0]) * 448);
218 fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
219 memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
220 fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
223 WINDOW_FUNC(eight_short)
225 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
226 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
227 const float *in = audio + 448;
228 float *out = sce->ret_buf;
231 for (w = 0; w < 8; w++) {
232 fdsp->vector_fmul (out, in, w ? pwindow : swindow, 128);
235 fdsp->vector_fmul_reverse(out, in, swindow, 128);
240 static void (*const apply_window[4])(AVFloatDSPContext *fdsp,
241 SingleChannelElement *sce,
242 const float *audio) = {
243 [ONLY_LONG_SEQUENCE] = apply_only_long_window,
244 [LONG_START_SEQUENCE] = apply_long_start_window,
245 [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
246 [LONG_STOP_SEQUENCE] = apply_long_stop_window
249 static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
253 float *output = sce->ret_buf;
255 apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);
257 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
258 s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
260 for (i = 0; i < 1024; i += 128)
261 s->mdct128.mdct_calc(&s->mdct128, sce->coeffs + i, output + i*2);
262 memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
263 memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));
267 * Encode ics_info element.
268 * @see Table 4.6 (syntax of ics_info)
270 static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
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
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]);
289 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
291 static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
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]);
303 * Produce integer coefficients from scalefactors provided by the model.
305 static void adjust_frame_information(ChannelElement *cpe, int chans)
308 int start, maxsfb, cmaxsfb;
310 for (ch = 0; ch < chans; ch++) {
311 IndividualChannelStream *ics = &cpe->ch[ch].ics;
314 cpe->ch[ch].pulse.num_pulse = 0;
315 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
316 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
317 start = (w+w2) * 128;
318 for (g = 0; g < ics->num_swb; g++) {
320 if (cpe->common_window && !ch && cpe->ms_mask[w*16 + g]) {
321 for (i = 0; i < ics->swb_sizes[g]; i++) {
322 cpe->ch[0].coeffs[start+i] = (cpe->ch[0].pcoeffs[start+i] + cpe->ch[1].pcoeffs[start+i]) * 0.5f;
323 cpe->ch[1].coeffs[start+i] = cpe->ch[0].coeffs[start+i] - cpe->ch[1].pcoeffs[start+i];
326 start += ics->swb_sizes[g];
328 for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
330 maxsfb = FFMAX(maxsfb, cmaxsfb);
333 ics->max_sfb = maxsfb;
335 //adjust zero bands for window groups
336 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
337 for (g = 0; g < ics->max_sfb; g++) {
339 for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
340 if (!cpe->ch[ch].zeroes[w2*16 + g]) {
345 cpe->ch[ch].zeroes[w*16 + g] = i;
350 if (chans > 1 && cpe->common_window) {
351 IndividualChannelStream *ics0 = &cpe->ch[0].ics;
352 IndividualChannelStream *ics1 = &cpe->ch[1].ics;
354 ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
355 ics1->max_sfb = ics0->max_sfb;
356 for (w = 0; w < ics0->num_windows*16; w += 16)
357 for (i = 0; i < ics0->max_sfb; i++)
358 if (cpe->ms_mask[w+i])
360 if (msc == 0 || ics0->max_sfb == 0)
363 cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
368 * Encode scalefactor band coding type.
370 static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
374 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
375 s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
379 * Encode scalefactors.
381 static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
382 SingleChannelElement *sce)
384 int off = sce->sf_idx[0], diff;
387 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
388 for (i = 0; i < sce->ics.max_sfb; i++) {
389 if (!sce->zeroes[w*16 + i]) {
390 diff = sce->sf_idx[w*16 + i] - off + SCALE_DIFF_ZERO;
391 av_assert0(diff >= 0 && diff <= 120);
392 off = sce->sf_idx[w*16 + i];
393 put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
402 static void encode_pulses(AACEncContext *s, Pulse *pulse)
406 put_bits(&s->pb, 1, !!pulse->num_pulse);
407 if (!pulse->num_pulse)
410 put_bits(&s->pb, 2, pulse->num_pulse - 1);
411 put_bits(&s->pb, 6, pulse->start);
412 for (i = 0; i < pulse->num_pulse; i++) {
413 put_bits(&s->pb, 5, pulse->pos[i]);
414 put_bits(&s->pb, 4, pulse->amp[i]);
419 * Encode spectral coefficients processed by psychoacoustic model.
421 static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
425 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
427 for (i = 0; i < sce->ics.max_sfb; i++) {
428 if (sce->zeroes[w*16 + i]) {
429 start += sce->ics.swb_sizes[i];
432 for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++)
433 s->coder->quantize_and_encode_band(s, &s->pb, sce->coeffs + start + w2*128,
434 sce->ics.swb_sizes[i],
435 sce->sf_idx[w*16 + i],
436 sce->band_type[w*16 + i],
438 start += sce->ics.swb_sizes[i];
444 * Encode one channel of audio data.
446 static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
447 SingleChannelElement *sce,
450 put_bits(&s->pb, 8, sce->sf_idx[0]);
452 put_ics_info(s, &sce->ics);
453 encode_band_info(s, sce);
454 encode_scale_factors(avctx, s, sce);
455 encode_pulses(s, &sce->pulse);
456 put_bits(&s->pb, 1, 0); //tns
457 put_bits(&s->pb, 1, 0); //ssr
458 encode_spectral_coeffs(s, sce);
463 * Write some auxiliary information about the created AAC file.
465 static void put_bitstream_info(AACEncContext *s, const char *name)
467 int i, namelen, padbits;
469 namelen = strlen(name) + 2;
470 put_bits(&s->pb, 3, TYPE_FIL);
471 put_bits(&s->pb, 4, FFMIN(namelen, 15));
473 put_bits(&s->pb, 8, namelen - 14);
474 put_bits(&s->pb, 4, 0); //extension type - filler
475 padbits = -put_bits_count(&s->pb) & 7;
476 avpriv_align_put_bits(&s->pb);
477 for (i = 0; i < namelen - 2; i++)
478 put_bits(&s->pb, 8, name[i]);
479 put_bits(&s->pb, 12 - padbits, 0);
483 * Copy input samples.
484 * Channels are reordered from libavcodec's default order to AAC order.
486 static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
489 int end = 2048 + (frame ? frame->nb_samples : 0);
490 const uint8_t *channel_map = aac_chan_maps[s->channels - 1];
492 /* copy and remap input samples */
493 for (ch = 0; ch < s->channels; ch++) {
494 /* copy last 1024 samples of previous frame to the start of the current frame */
495 memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
497 /* copy new samples and zero any remaining samples */
499 memcpy(&s->planar_samples[ch][2048],
500 frame->extended_data[channel_map[ch]],
501 frame->nb_samples * sizeof(s->planar_samples[0][0]));
503 memset(&s->planar_samples[ch][end], 0,
504 (3072 - end) * sizeof(s->planar_samples[0][0]));
508 static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
509 const AVFrame *frame, int *got_packet_ptr)
511 AACEncContext *s = avctx->priv_data;
512 float **samples = s->planar_samples, *samples2, *la, *overlap;
514 int i, ch, w, g, chans, tag, start_ch, ret, ms_mode = 0;
515 int chan_el_counter[4];
516 FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
518 if (s->last_frame == 2)
521 /* add current frame to queue */
523 if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
527 copy_input_samples(s, frame);
529 ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
531 if (!avctx->frame_number)
535 for (i = 0; i < s->chan_map[0]; i++) {
536 FFPsyWindowInfo* wi = windows + start_ch;
537 tag = s->chan_map[i+1];
538 chans = tag == TYPE_CPE ? 2 : 1;
540 for (ch = 0; ch < chans; ch++) {
541 IndividualChannelStream *ics = &cpe->ch[ch].ics;
542 int cur_channel = start_ch + ch;
543 overlap = &samples[cur_channel][0];
544 samples2 = overlap + 1024;
545 la = samples2 + (448+64);
548 if (tag == TYPE_LFE) {
549 wi[ch].window_type[0] = ONLY_LONG_SEQUENCE;
550 wi[ch].window_shape = 0;
551 wi[ch].num_windows = 1;
552 wi[ch].grouping[0] = 1;
554 /* Only the lowest 12 coefficients are used in a LFE channel.
555 * The expression below results in only the bottom 8 coefficients
556 * being used for 11.025kHz to 16kHz sample rates.
558 ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
560 wi[ch] = s->psy.model->window(&s->psy, samples2, la, cur_channel,
561 ics->window_sequence[0]);
563 ics->window_sequence[1] = ics->window_sequence[0];
564 ics->window_sequence[0] = wi[ch].window_type[0];
565 ics->use_kb_window[1] = ics->use_kb_window[0];
566 ics->use_kb_window[0] = wi[ch].window_shape;
567 ics->num_windows = wi[ch].num_windows;
568 ics->swb_sizes = s->psy.bands [ics->num_windows == 8];
569 ics->num_swb = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
570 for (w = 0; w < ics->num_windows; w++)
571 ics->group_len[w] = wi[ch].grouping[w];
573 apply_window_and_mdct(s, &cpe->ch[ch], overlap);
574 if (isnan(cpe->ch->coeffs[0])) {
575 av_log(avctx, AV_LOG_ERROR, "Input contains NaN\n");
576 return AVERROR(EINVAL);
581 if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels)) < 0)
586 init_put_bits(&s->pb, avpkt->data, avpkt->size);
588 if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & CODEC_FLAG_BITEXACT))
589 put_bitstream_info(s, LIBAVCODEC_IDENT);
591 memset(chan_el_counter, 0, sizeof(chan_el_counter));
592 for (i = 0; i < s->chan_map[0]; i++) {
593 FFPsyWindowInfo* wi = windows + start_ch;
594 const float *coeffs[2];
595 tag = s->chan_map[i+1];
596 chans = tag == TYPE_CPE ? 2 : 1;
598 put_bits(&s->pb, 3, tag);
599 put_bits(&s->pb, 4, chan_el_counter[tag]++);
600 for (ch = 0; ch < chans; ch++)
601 coeffs[ch] = cpe->ch[ch].coeffs;
602 s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
603 for (ch = 0; ch < chans; ch++) {
604 s->cur_channel = start_ch + ch;
605 s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
607 cpe->common_window = 0;
609 && wi[0].window_type[0] == wi[1].window_type[0]
610 && wi[0].window_shape == wi[1].window_shape) {
612 cpe->common_window = 1;
613 for (w = 0; w < wi[0].num_windows; w++) {
614 if (wi[0].grouping[w] != wi[1].grouping[w]) {
615 cpe->common_window = 0;
620 s->cur_channel = start_ch;
621 if (s->options.stereo_mode && cpe->common_window) {
622 if (s->options.stereo_mode > 0) {
623 IndividualChannelStream *ics = &cpe->ch[0].ics;
624 for (w = 0; w < ics->num_windows; w += ics->group_len[w])
625 for (g = 0; g < ics->num_swb; g++)
626 cpe->ms_mask[w*16+g] = 1;
627 } else if (s->coder->search_for_ms) {
628 s->coder->search_for_ms(s, cpe, s->lambda);
631 adjust_frame_information(cpe, chans);
633 put_bits(&s->pb, 1, cpe->common_window);
634 if (cpe->common_window) {
635 put_ics_info(s, &cpe->ch[0].ics);
636 encode_ms_info(&s->pb, cpe);
637 if (cpe->ms_mode) ms_mode = 1;
640 for (ch = 0; ch < chans; ch++) {
641 s->cur_channel = start_ch + ch;
642 encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
647 frame_bits = put_bits_count(&s->pb);
648 if (frame_bits <= 6144 * s->channels - 3) {
649 s->psy.bitres.bits = frame_bits / s->channels;
653 for (i = 0; i < s->chan_map[0]; i++) {
654 // Must restore coeffs
655 chans = tag == TYPE_CPE ? 2 : 1;
657 for (ch = 0; ch < chans; ch++)
658 memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
662 s->lambda *= avctx->bit_rate * 1024.0f / avctx->sample_rate / frame_bits;
666 put_bits(&s->pb, 3, TYPE_END);
667 flush_put_bits(&s->pb);
668 avctx->frame_bits = put_bits_count(&s->pb);
670 // rate control stuff
671 if (!(avctx->flags & CODEC_FLAG_QSCALE)) {
672 float ratio = avctx->bit_rate * 1024.0f / avctx->sample_rate / avctx->frame_bits;
674 s->lambda = FFMIN(s->lambda, 65536.f);
680 ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
683 avpkt->size = put_bits_count(&s->pb) >> 3;
688 static av_cold int aac_encode_end(AVCodecContext *avctx)
690 AACEncContext *s = avctx->priv_data;
692 ff_mdct_end(&s->mdct1024);
693 ff_mdct_end(&s->mdct128);
696 ff_psy_preprocess_end(s->psypp);
697 av_freep(&s->buffer.samples);
700 ff_af_queue_close(&s->afq);
704 static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
708 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & CODEC_FLAG_BITEXACT);
710 return AVERROR(ENOMEM);
713 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
714 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
715 ff_init_ff_sine_windows(10);
716 ff_init_ff_sine_windows(7);
718 if (ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0))
720 if (ret = ff_mdct_init(&s->mdct128, 8, 0, 32768.0))
726 static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
729 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->buffer.samples, s->channels, 3 * 1024 * sizeof(s->buffer.samples[0]), alloc_fail);
730 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->cpe, s->chan_map[0], sizeof(ChannelElement), alloc_fail);
731 FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + FF_INPUT_BUFFER_PADDING_SIZE, alloc_fail);
733 for(ch = 0; ch < s->channels; ch++)
734 s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
738 return AVERROR(ENOMEM);
741 static av_cold int aac_encode_init(AVCodecContext *avctx)
743 AACEncContext *s = avctx->priv_data;
745 const uint8_t *sizes[2];
746 uint8_t grouping[AAC_MAX_CHANNELS];
749 avctx->frame_size = 1024;
751 for (i = 0; i < 16; i++)
752 if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
755 s->channels = avctx->channels;
758 "Unsupported sample rate %d\n", avctx->sample_rate);
759 ERROR_IF(s->channels > AAC_MAX_CHANNELS,
760 "Unsupported number of channels: %d\n", s->channels);
761 ERROR_IF(avctx->profile != FF_PROFILE_UNKNOWN && avctx->profile != FF_PROFILE_AAC_LOW,
762 "Unsupported profile %d\n", avctx->profile);
763 ERROR_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
764 "Too many bits per frame requested\n");
766 s->samplerate_index = i;
768 s->chan_map = aac_chan_configs[s->channels-1];
770 if ((ret = dsp_init(avctx, s)) < 0)
773 if ((ret = alloc_buffers(avctx, s)) < 0)
776 avctx->extradata_size = 5;
777 put_audio_specific_config(avctx);
779 sizes[0] = swb_size_1024[i];
780 sizes[1] = swb_size_128[i];
781 lengths[0] = ff_aac_num_swb_1024[i];
782 lengths[1] = ff_aac_num_swb_128[i];
783 for (i = 0; i < s->chan_map[0]; i++)
784 grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
785 if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
786 s->chan_map[0], grouping)) < 0)
788 s->psypp = ff_psy_preprocess_init(avctx);
789 s->coder = &ff_aac_coders[s->options.aac_coder];
792 ff_aac_coder_init_mips(s);
794 s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
798 for (i = 0; i < 428; i++)
799 ff_aac_pow34sf_tab[i] = sqrt(ff_aac_pow2sf_tab[i] * sqrt(ff_aac_pow2sf_tab[i]));
801 avctx->initial_padding = 1024;
802 ff_af_queue_init(avctx, &s->afq);
806 aac_encode_end(avctx);
810 #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
811 static const AVOption aacenc_options[] = {
812 {"stereo_mode", "Stereo coding method", offsetof(AACEncContext, options.stereo_mode), AV_OPT_TYPE_INT, {.i64 = 0}, -1, 1, AACENC_FLAGS, "stereo_mode"},
813 {"auto", "Selected by the Encoder", 0, AV_OPT_TYPE_CONST, {.i64 = -1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
814 {"ms_off", "Disable Mid/Side coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
815 {"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"},
816 {"aac_coder", "", offsetof(AACEncContext, options.aac_coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_TWOLOOP}, 0, AAC_CODER_NB-1, AACENC_FLAGS, "aac_coder"},
817 {"faac", "FAAC-inspired method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAAC}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
818 {"anmr", "ANMR method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
819 {"twoloop", "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
820 {"fast", "Constant quantizer", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
824 static const AVClass aacenc_class = {
826 av_default_item_name,
828 LIBAVUTIL_VERSION_INT,
831 /* duplicated from avpriv_mpeg4audio_sample_rates to avoid shared build
833 static const int mpeg4audio_sample_rates[16] = {
834 96000, 88200, 64000, 48000, 44100, 32000,
835 24000, 22050, 16000, 12000, 11025, 8000, 7350
838 AVCodec ff_aac_encoder = {
840 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
841 .type = AVMEDIA_TYPE_AUDIO,
842 .id = AV_CODEC_ID_AAC,
843 .priv_data_size = sizeof(AACEncContext),
844 .init = aac_encode_init,
845 .encode2 = aac_encode_frame,
846 .close = aac_encode_end,
847 .supported_samplerates = mpeg4audio_sample_rates,
848 .capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY |
849 CODEC_CAP_EXPERIMENTAL,
850 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
851 AV_SAMPLE_FMT_NONE },
852 .priv_class = &aacenc_class,