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 ***********************************/
32 #include "libavutil/float_dsp.h"
33 #include "libavutil/opt.h"
37 #include "mpeg4audio.h"
44 #include "aacenctab.h"
45 #include "aacenc_utils.h"
50 * Make AAC audio config object.
51 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
53 static void put_audio_specific_config(AVCodecContext *avctx)
56 AACEncContext *s = avctx->priv_data;
58 init_put_bits(&pb, avctx->extradata, avctx->extradata_size);
59 put_bits(&pb, 5, s->profile+1); //profile
60 put_bits(&pb, 4, s->samplerate_index); //sample rate index
61 put_bits(&pb, 4, s->channels);
63 put_bits(&pb, 1, 0); //frame length - 1024 samples
64 put_bits(&pb, 1, 0); //does not depend on core coder
65 put_bits(&pb, 1, 0); //is not extension
67 //Explicitly Mark SBR absent
68 put_bits(&pb, 11, 0x2b7); //sync extension
69 put_bits(&pb, 5, AOT_SBR);
74 #define WINDOW_FUNC(type) \
75 static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \
76 SingleChannelElement *sce, \
79 WINDOW_FUNC(only_long)
81 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
82 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
83 float *out = sce->ret_buf;
85 fdsp->vector_fmul (out, audio, lwindow, 1024);
86 fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
89 WINDOW_FUNC(long_start)
91 const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
92 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
93 float *out = sce->ret_buf;
95 fdsp->vector_fmul(out, audio, lwindow, 1024);
96 memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
97 fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
98 memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
101 WINDOW_FUNC(long_stop)
103 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
104 const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
105 float *out = sce->ret_buf;
107 memset(out, 0, sizeof(out[0]) * 448);
108 fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
109 memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
110 fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
113 WINDOW_FUNC(eight_short)
115 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
116 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
117 const float *in = audio + 448;
118 float *out = sce->ret_buf;
121 for (w = 0; w < 8; w++) {
122 fdsp->vector_fmul (out, in, w ? pwindow : swindow, 128);
125 fdsp->vector_fmul_reverse(out, in, swindow, 128);
130 static void (*const apply_window[4])(AVFloatDSPContext *fdsp,
131 SingleChannelElement *sce,
132 const float *audio) = {
133 [ONLY_LONG_SEQUENCE] = apply_only_long_window,
134 [LONG_START_SEQUENCE] = apply_long_start_window,
135 [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
136 [LONG_STOP_SEQUENCE] = apply_long_stop_window
139 static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
143 float *output = sce->ret_buf;
145 apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);
147 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
148 s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
150 for (i = 0; i < 1024; i += 128)
151 s->mdct128.mdct_calc(&s->mdct128, &sce->coeffs[i], output + i*2);
152 memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
153 memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));
157 * Encode ics_info element.
158 * @see Table 4.6 (syntax of ics_info)
160 static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
164 put_bits(&s->pb, 1, 0); // ics_reserved bit
165 put_bits(&s->pb, 2, info->window_sequence[0]);
166 put_bits(&s->pb, 1, info->use_kb_window[0]);
167 if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
168 put_bits(&s->pb, 6, info->max_sfb);
169 put_bits(&s->pb, 1, !!info->predictor_present);
171 put_bits(&s->pb, 4, info->max_sfb);
172 for (w = 1; w < 8; w++)
173 put_bits(&s->pb, 1, !info->group_len[w]);
179 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
181 static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
185 put_bits(pb, 2, cpe->ms_mode);
186 if (cpe->ms_mode == 1)
187 for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
188 for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
189 put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
193 * Produce integer coefficients from scalefactors provided by the model.
195 static void adjust_frame_information(ChannelElement *cpe, int chans)
200 for (ch = 0; ch < chans; ch++) {
201 IndividualChannelStream *ics = &cpe->ch[ch].ics;
203 cpe->ch[ch].pulse.num_pulse = 0;
204 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
205 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
206 for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
208 maxsfb = FFMAX(maxsfb, cmaxsfb);
211 ics->max_sfb = maxsfb;
213 //adjust zero bands for window groups
214 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
215 for (g = 0; g < ics->max_sfb; g++) {
217 for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
218 if (!cpe->ch[ch].zeroes[w2*16 + g]) {
223 cpe->ch[ch].zeroes[w*16 + g] = i;
228 if (chans > 1 && cpe->common_window) {
229 IndividualChannelStream *ics0 = &cpe->ch[0].ics;
230 IndividualChannelStream *ics1 = &cpe->ch[1].ics;
232 ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
233 ics1->max_sfb = ics0->max_sfb;
234 for (w = 0; w < ics0->num_windows*16; w += 16)
235 for (i = 0; i < ics0->max_sfb; i++)
236 if (cpe->ms_mask[w+i])
238 if (msc == 0 || ics0->max_sfb == 0)
241 cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
245 static void apply_intensity_stereo(ChannelElement *cpe)
248 IndividualChannelStream *ics = &cpe->ch[0].ics;
249 if (!cpe->common_window)
251 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
252 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
253 int start = (w+w2) * 128;
254 for (g = 0; g < ics->num_swb; g++) {
255 int p = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
256 float scale = cpe->ch[0].is_ener[w*16+g];
257 if (!cpe->is_mask[w*16 + g]) {
258 start += ics->swb_sizes[g];
261 for (i = 0; i < ics->swb_sizes[g]; i++) {
262 float sum = (cpe->ch[0].coeffs[start+i] + p*cpe->ch[1].coeffs[start+i])*scale;
263 cpe->ch[0].coeffs[start+i] = sum;
264 cpe->ch[1].coeffs[start+i] = 0.0f;
266 start += ics->swb_sizes[g];
272 static void apply_mid_side_stereo(ChannelElement *cpe)
275 IndividualChannelStream *ics = &cpe->ch[0].ics;
276 if (!cpe->common_window)
278 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
279 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
280 int start = (w+w2) * 128;
281 for (g = 0; g < ics->num_swb; g++) {
282 if (!cpe->ms_mask[w*16 + g]) {
283 start += ics->swb_sizes[g];
286 for (i = 0; i < ics->swb_sizes[g]; i++) {
287 float L = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) * 0.5f;
288 float R = L - cpe->ch[1].coeffs[start+i];
289 cpe->ch[0].coeffs[start+i] = L;
290 cpe->ch[1].coeffs[start+i] = R;
292 start += ics->swb_sizes[g];
299 * Encode scalefactor band coding type.
301 static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
305 if (s->coder->set_special_band_scalefactors)
306 s->coder->set_special_band_scalefactors(s, sce);
308 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
309 s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
313 * Encode scalefactors.
315 static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
316 SingleChannelElement *sce)
318 int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;
319 int off_is = 0, noise_flag = 1;
322 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
323 for (i = 0; i < sce->ics.max_sfb; i++) {
324 if (!sce->zeroes[w*16 + i]) {
325 if (sce->band_type[w*16 + i] == NOISE_BT) {
326 diff = sce->sf_idx[w*16 + i] - off_pns;
327 off_pns = sce->sf_idx[w*16 + i];
328 if (noise_flag-- > 0) {
329 put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);
332 } else if (sce->band_type[w*16 + i] == INTENSITY_BT ||
333 sce->band_type[w*16 + i] == INTENSITY_BT2) {
334 diff = sce->sf_idx[w*16 + i] - off_is;
335 off_is = sce->sf_idx[w*16 + i];
337 diff = sce->sf_idx[w*16 + i] - off_sf;
338 off_sf = sce->sf_idx[w*16 + i];
340 diff += SCALE_DIFF_ZERO;
341 av_assert0(diff >= 0 && diff <= 120);
342 put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
351 static void encode_pulses(AACEncContext *s, Pulse *pulse)
355 put_bits(&s->pb, 1, !!pulse->num_pulse);
356 if (!pulse->num_pulse)
359 put_bits(&s->pb, 2, pulse->num_pulse - 1);
360 put_bits(&s->pb, 6, pulse->start);
361 for (i = 0; i < pulse->num_pulse; i++) {
362 put_bits(&s->pb, 5, pulse->pos[i]);
363 put_bits(&s->pb, 4, pulse->amp[i]);
368 * Encode spectral coefficients processed by psychoacoustic model.
370 static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
374 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
376 for (i = 0; i < sce->ics.max_sfb; i++) {
377 if (sce->zeroes[w*16 + i]) {
378 start += sce->ics.swb_sizes[i];
381 for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) {
382 s->coder->quantize_and_encode_band(s, &s->pb,
383 &sce->coeffs[start + w2*128],
384 NULL, sce->ics.swb_sizes[i],
385 sce->sf_idx[w*16 + i],
386 sce->band_type[w*16 + i],
388 sce->ics.window_clipping[w]);
390 start += sce->ics.swb_sizes[i];
396 * Downscale spectral coefficients for near-clipping windows to avoid artifacts
398 static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)
402 if (sce->ics.clip_avoidance_factor < 1.0f) {
403 for (w = 0; w < sce->ics.num_windows; w++) {
405 for (i = 0; i < sce->ics.max_sfb; i++) {
406 float *swb_coeffs = &sce->coeffs[start + w*128];
407 for (j = 0; j < sce->ics.swb_sizes[i]; j++)
408 swb_coeffs[j] *= sce->ics.clip_avoidance_factor;
409 start += sce->ics.swb_sizes[i];
416 * Encode one channel of audio data.
418 static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
419 SingleChannelElement *sce,
422 put_bits(&s->pb, 8, sce->sf_idx[0]);
423 if (!common_window) {
424 put_ics_info(s, &sce->ics);
425 if (s->coder->encode_main_pred)
426 s->coder->encode_main_pred(s, sce);
428 encode_band_info(s, sce);
429 encode_scale_factors(avctx, s, sce);
430 encode_pulses(s, &sce->pulse);
431 put_bits(&s->pb, 1, !!sce->tns.present);
432 if (s->coder->encode_tns_info)
433 s->coder->encode_tns_info(s, sce);
434 put_bits(&s->pb, 1, 0); //ssr
435 encode_spectral_coeffs(s, sce);
440 * Write some auxiliary information about the created AAC file.
442 static void put_bitstream_info(AACEncContext *s, const char *name)
444 int i, namelen, padbits;
446 namelen = strlen(name) + 2;
447 put_bits(&s->pb, 3, TYPE_FIL);
448 put_bits(&s->pb, 4, FFMIN(namelen, 15));
450 put_bits(&s->pb, 8, namelen - 14);
451 put_bits(&s->pb, 4, 0); //extension type - filler
452 padbits = -put_bits_count(&s->pb) & 7;
453 avpriv_align_put_bits(&s->pb);
454 for (i = 0; i < namelen - 2; i++)
455 put_bits(&s->pb, 8, name[i]);
456 put_bits(&s->pb, 12 - padbits, 0);
460 * Copy input samples.
461 * Channels are reordered from libavcodec's default order to AAC order.
463 static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
466 int end = 2048 + (frame ? frame->nb_samples : 0);
467 const uint8_t *channel_map = aac_chan_maps[s->channels - 1];
469 /* copy and remap input samples */
470 for (ch = 0; ch < s->channels; ch++) {
471 /* copy last 1024 samples of previous frame to the start of the current frame */
472 memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
474 /* copy new samples and zero any remaining samples */
476 memcpy(&s->planar_samples[ch][2048],
477 frame->extended_data[channel_map[ch]],
478 frame->nb_samples * sizeof(s->planar_samples[0][0]));
480 memset(&s->planar_samples[ch][end], 0,
481 (3072 - end) * sizeof(s->planar_samples[0][0]));
485 static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
486 const AVFrame *frame, int *got_packet_ptr)
488 AACEncContext *s = avctx->priv_data;
489 float **samples = s->planar_samples, *samples2, *la, *overlap;
491 SingleChannelElement *sce;
492 int i, ch, w, chans, tag, start_ch, ret;
493 int ms_mode = 0, is_mode = 0, tns_mode = 0, pred_mode = 0;
494 int chan_el_counter[4];
495 FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
497 if (s->last_frame == 2)
500 /* add current frame to queue */
502 if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
506 copy_input_samples(s, frame);
508 ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
510 if (!avctx->frame_number)
514 for (i = 0; i < s->chan_map[0]; i++) {
515 FFPsyWindowInfo* wi = windows + start_ch;
516 tag = s->chan_map[i+1];
517 chans = tag == TYPE_CPE ? 2 : 1;
519 for (ch = 0; ch < chans; ch++) {
520 IndividualChannelStream *ics = &cpe->ch[ch].ics;
521 int cur_channel = start_ch + ch;
522 float clip_avoidance_factor;
523 overlap = &samples[cur_channel][0];
524 samples2 = overlap + 1024;
525 la = samples2 + (448+64);
528 if (tag == TYPE_LFE) {
529 wi[ch].window_type[0] = ONLY_LONG_SEQUENCE;
530 wi[ch].window_shape = 0;
531 wi[ch].num_windows = 1;
532 wi[ch].grouping[0] = 1;
534 /* Only the lowest 12 coefficients are used in a LFE channel.
535 * The expression below results in only the bottom 8 coefficients
536 * being used for 11.025kHz to 16kHz sample rates.
538 ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
540 wi[ch] = s->psy.model->window(&s->psy, samples2, la, cur_channel,
541 ics->window_sequence[0]);
543 ics->window_sequence[1] = ics->window_sequence[0];
544 ics->window_sequence[0] = wi[ch].window_type[0];
545 ics->use_kb_window[1] = ics->use_kb_window[0];
546 ics->use_kb_window[0] = wi[ch].window_shape;
547 ics->num_windows = wi[ch].num_windows;
548 ics->swb_sizes = s->psy.bands [ics->num_windows == 8];
549 ics->num_swb = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
550 ics->swb_offset = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
551 ff_swb_offset_128 [s->samplerate_index]:
552 ff_swb_offset_1024[s->samplerate_index];
553 ics->tns_max_bands = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
554 ff_tns_max_bands_128 [s->samplerate_index]:
555 ff_tns_max_bands_1024[s->samplerate_index];
556 clip_avoidance_factor = 0.0f;
557 for (w = 0; w < ics->num_windows; w++)
558 ics->group_len[w] = wi[ch].grouping[w];
559 for (w = 0; w < ics->num_windows; w++) {
560 if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
561 ics->window_clipping[w] = 1;
562 clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
564 ics->window_clipping[w] = 0;
567 if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
568 ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
570 ics->clip_avoidance_factor = 1.0f;
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);
578 avoid_clipping(s, &cpe->ch[ch]);
582 if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)
587 init_put_bits(&s->pb, avpkt->data, avpkt->size);
589 if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
590 put_bitstream_info(s, LIBAVCODEC_IDENT);
592 memset(chan_el_counter, 0, sizeof(chan_el_counter));
593 for (i = 0; i < s->chan_map[0]; i++) {
594 FFPsyWindowInfo* wi = windows + start_ch;
595 const float *coeffs[2];
596 tag = s->chan_map[i+1];
597 chans = tag == TYPE_CPE ? 2 : 1;
599 cpe->common_window = 0;
600 memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
601 memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
602 put_bits(&s->pb, 3, tag);
603 put_bits(&s->pb, 4, chan_el_counter[tag]++);
604 for (ch = 0; ch < chans; ch++) {
606 coeffs[ch] = sce->coeffs;
607 sce->ics.predictor_present = 0;
608 memset(&sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));
609 memset(&sce->tns, 0, sizeof(TemporalNoiseShaping));
610 for (w = 0; w < 128; w++)
611 if (sce->band_type[w] > RESERVED_BT)
612 sce->band_type[w] = 0;
614 s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
615 for (ch = 0; ch < chans; ch++) {
616 s->cur_channel = start_ch + ch;
617 s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
620 && wi[0].window_type[0] == wi[1].window_type[0]
621 && wi[0].window_shape == wi[1].window_shape) {
623 cpe->common_window = 1;
624 for (w = 0; w < wi[0].num_windows; w++) {
625 if (wi[0].grouping[w] != wi[1].grouping[w]) {
626 cpe->common_window = 0;
631 for (ch = 0; ch < chans; ch++) { /* TNS and PNS */
633 s->cur_channel = start_ch + ch;
634 if (s->options.pns && s->coder->search_for_pns)
635 s->coder->search_for_pns(s, avctx, sce);
636 if (s->options.tns && s->coder->search_for_tns)
637 s->coder->search_for_tns(s, sce);
638 if (s->options.tns && s->coder->apply_tns_filt)
639 s->coder->apply_tns_filt(s, sce);
640 if (sce->tns.present)
643 s->cur_channel = start_ch;
644 if (s->options.intensity_stereo) { /* Intensity Stereo */
645 if (s->coder->search_for_is)
646 s->coder->search_for_is(s, avctx, cpe);
647 if (cpe->is_mode) is_mode = 1;
648 apply_intensity_stereo(cpe);
650 if (s->options.pred) { /* Prediction */
651 for (ch = 0; ch < chans; ch++) {
653 s->cur_channel = start_ch + ch;
654 if (s->options.pred && s->coder->search_for_pred)
655 s->coder->search_for_pred(s, sce);
656 if (cpe->ch[ch].ics.predictor_present) pred_mode = 1;
658 if (s->coder->adjust_common_prediction)
659 s->coder->adjust_common_prediction(s, cpe);
660 for (ch = 0; ch < chans; ch++) {
662 s->cur_channel = start_ch + ch;
663 if (s->options.pred && s->coder->apply_main_pred)
664 s->coder->apply_main_pred(s, sce);
666 s->cur_channel = start_ch;
668 if (s->options.stereo_mode) { /* Mid/Side stereo */
669 if (s->options.stereo_mode == -1 && s->coder->search_for_ms)
670 s->coder->search_for_ms(s, cpe);
671 else if (cpe->common_window)
672 memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask));
673 for (w = 0; w < 128; w++)
674 cpe->ms_mask[w] = cpe->is_mask[w] ? 0 : cpe->ms_mask[w];
675 apply_mid_side_stereo(cpe);
677 adjust_frame_information(cpe, chans);
679 put_bits(&s->pb, 1, cpe->common_window);
680 if (cpe->common_window) {
681 put_ics_info(s, &cpe->ch[0].ics);
682 if (s->coder->encode_main_pred)
683 s->coder->encode_main_pred(s, &cpe->ch[0]);
684 encode_ms_info(&s->pb, cpe);
685 if (cpe->ms_mode) ms_mode = 1;
688 for (ch = 0; ch < chans; ch++) {
689 s->cur_channel = start_ch + ch;
690 encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
695 frame_bits = put_bits_count(&s->pb);
696 if (frame_bits <= 6144 * s->channels - 3) {
697 s->psy.bitres.bits = frame_bits / s->channels;
700 if (is_mode || ms_mode || tns_mode || pred_mode) {
701 for (i = 0; i < s->chan_map[0]; i++) {
702 // Must restore coeffs
703 chans = tag == TYPE_CPE ? 2 : 1;
705 for (ch = 0; ch < chans; ch++)
706 memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
710 s->lambda *= avctx->bit_rate * 1024.0f / avctx->sample_rate / frame_bits;
714 put_bits(&s->pb, 3, TYPE_END);
715 flush_put_bits(&s->pb);
716 avctx->frame_bits = put_bits_count(&s->pb);
718 // rate control stuff
719 if (!(avctx->flags & AV_CODEC_FLAG_QSCALE)) {
720 float ratio = avctx->bit_rate * 1024.0f / avctx->sample_rate / avctx->frame_bits;
722 s->lambda = FFMIN(s->lambda, 65536.f);
728 ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
731 avpkt->size = put_bits_count(&s->pb) >> 3;
736 static av_cold int aac_encode_end(AVCodecContext *avctx)
738 AACEncContext *s = avctx->priv_data;
740 ff_mdct_end(&s->mdct1024);
741 ff_mdct_end(&s->mdct128);
745 ff_psy_preprocess_end(s->psypp);
746 av_freep(&s->buffer.samples);
749 ff_af_queue_close(&s->afq);
753 static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
757 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
759 return AVERROR(ENOMEM);
762 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
763 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
764 ff_init_ff_sine_windows(10);
765 ff_init_ff_sine_windows(7);
767 if ((ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) < 0)
769 if ((ret = ff_mdct_init(&s->mdct128, 8, 0, 32768.0)) < 0)
775 static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
778 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->buffer.samples, s->channels, 3 * 1024 * sizeof(s->buffer.samples[0]), alloc_fail);
779 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->cpe, s->chan_map[0], sizeof(ChannelElement), alloc_fail);
780 FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + AV_INPUT_BUFFER_PADDING_SIZE, alloc_fail);
782 for(ch = 0; ch < s->channels; ch++)
783 s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
787 return AVERROR(ENOMEM);
790 static av_cold int aac_encode_init(AVCodecContext *avctx)
792 AACEncContext *s = avctx->priv_data;
794 const uint8_t *sizes[2];
795 uint8_t grouping[AAC_MAX_CHANNELS];
798 avctx->frame_size = 1024;
800 for (i = 0; i < 16; i++)
801 if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
804 s->channels = avctx->channels;
806 ERROR_IF(i == 16 || i >= ff_aac_swb_size_1024_len || i >= ff_aac_swb_size_128_len,
807 "Unsupported sample rate %d\n", avctx->sample_rate);
808 ERROR_IF(s->channels > AAC_MAX_CHANNELS,
809 "Unsupported number of channels: %d\n", s->channels);
810 WARN_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
811 "Too many bits per frame requested, clamping to max\n");
812 if (avctx->profile == FF_PROFILE_AAC_MAIN) {
814 } else if ((avctx->profile == FF_PROFILE_AAC_LOW ||
815 avctx->profile == FF_PROFILE_UNKNOWN) && s->options.pred) {
816 s->profile = 0; /* Main */
817 WARN_IF(1, "Prediction requested, changing profile to AAC-Main\n");
818 } else if (avctx->profile == FF_PROFILE_AAC_LOW ||
819 avctx->profile == FF_PROFILE_UNKNOWN) {
820 s->profile = 1; /* Low */
822 ERROR_IF(1, "Unsupported profile %d\n", avctx->profile);
825 if (s->options.aac_coder != AAC_CODER_TWOLOOP) {
826 s->options.intensity_stereo = 0;
830 avctx->bit_rate = (int)FFMIN(
831 6144 * s->channels / 1024.0 * avctx->sample_rate,
834 s->samplerate_index = i;
836 s->chan_map = aac_chan_configs[s->channels-1];
838 if ((ret = dsp_init(avctx, s)) < 0)
841 if ((ret = alloc_buffers(avctx, s)) < 0)
844 avctx->extradata_size = 5;
845 put_audio_specific_config(avctx);
847 sizes[0] = ff_aac_swb_size_1024[i];
848 sizes[1] = ff_aac_swb_size_128[i];
849 lengths[0] = ff_aac_num_swb_1024[i];
850 lengths[1] = ff_aac_num_swb_128[i];
851 for (i = 0; i < s->chan_map[0]; i++)
852 grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
853 if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
854 s->chan_map[0], grouping)) < 0)
856 s->psypp = ff_psy_preprocess_init(avctx);
857 s->coder = &ff_aac_coders[s->options.aac_coder];
858 ff_lpc_init(&s->lpc, 2*avctx->frame_size, TNS_MAX_ORDER, FF_LPC_TYPE_LEVINSON);
861 ff_aac_coder_init_mips(s);
863 s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
867 avctx->initial_padding = 1024;
868 ff_af_queue_init(avctx, &s->afq);
872 aac_encode_end(avctx);
876 #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
877 static const AVOption aacenc_options[] = {
878 {"stereo_mode", "Stereo coding method", offsetof(AACEncContext, options.stereo_mode), AV_OPT_TYPE_INT, {.i64 = 0}, -1, 1, AACENC_FLAGS, "stereo_mode"},
879 {"auto", "Selected by the Encoder", 0, AV_OPT_TYPE_CONST, {.i64 = -1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
880 {"ms_off", "Disable Mid/Side coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
881 {"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"},
882 {"aac_coder", "Coding algorithm", offsetof(AACEncContext, options.aac_coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_TWOLOOP}, 0, AAC_CODER_NB-1, AACENC_FLAGS, "aac_coder"},
883 {"faac", "FAAC-inspired method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAAC}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
884 {"anmr", "ANMR method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
885 {"twoloop", "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
886 {"fast", "Constant quantizer", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
887 {"aac_pns", "Perceptual Noise Substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_INT, {.i64 = 1}, 0, 1, AACENC_FLAGS, "aac_pns"},
888 {"disable", "Disable perceptual noise substitution", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pns"},
889 {"enable", "Enable perceptual noise substitution", 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pns"},
890 {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_INT, {.i64 = 1}, 0, 1, AACENC_FLAGS, "intensity_stereo"},
891 {"disable", "Disable intensity stereo coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0}, INT_MIN, INT_MAX, AACENC_FLAGS, "intensity_stereo"},
892 {"enable", "Enable intensity stereo coding", 0, AV_OPT_TYPE_CONST, {.i64 = 1}, INT_MIN, INT_MAX, AACENC_FLAGS, "intensity_stereo"},
893 {"aac_tns", "Temporal noise shaping", offsetof(AACEncContext, options.tns), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, AACENC_FLAGS, "aac_tns"},
894 {"disable", "Disable temporal noise shaping", 0, AV_OPT_TYPE_CONST, {.i64 = 0}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_tns"},
895 {"enable", "Enable temporal noise shaping", 0, AV_OPT_TYPE_CONST, {.i64 = 1}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_tns"},
896 {"aac_pred", "AAC-Main prediction", offsetof(AACEncContext, options.pred), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, AACENC_FLAGS, "aac_pred"},
897 {"disable", "Disable AAC-Main prediction", 0, AV_OPT_TYPE_CONST, {.i64 = 0}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pred"},
898 {"enable", "Enable AAC-Main prediction", 0, AV_OPT_TYPE_CONST, {.i64 = 1}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pred"},
902 static const AVClass aacenc_class = {
904 av_default_item_name,
906 LIBAVUTIL_VERSION_INT,
909 AVCodec ff_aac_encoder = {
911 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
912 .type = AVMEDIA_TYPE_AUDIO,
913 .id = AV_CODEC_ID_AAC,
914 .priv_data_size = sizeof(AACEncContext),
915 .init = aac_encode_init,
916 .encode2 = aac_encode_frame,
917 .close = aac_encode_end,
918 .supported_samplerates = mpeg4audio_sample_rates,
919 .capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY |
920 AV_CODEC_CAP_EXPERIMENTAL,
921 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
922 AV_SAMPLE_FMT_NONE },
923 .priv_class = &aacenc_class,