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)
199 IndividualChannelStream *ics;
201 if (cpe->common_window) {
202 ics = &cpe->ch[0].ics;
203 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
204 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
205 int start = (w+w2) * 128;
206 for (g = 0; g < ics->num_swb; g++) {
207 //apply Intensity stereo coeffs transformation
208 if (cpe->is_mask[w*16 + g]) {
209 int p = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
210 float scale = cpe->ch[0].is_ener[w*16+g];
211 for (i = 0; i < ics->swb_sizes[g]; i++) {
212 cpe->ch[0].coeffs[start+i] = (cpe->ch[0].coeffs[start+i] + p*cpe->ch[1].coeffs[start+i]) * scale;
213 cpe->ch[1].coeffs[start+i] = 0.0f;
215 } else if (cpe->ms_mask[w*16 + g] &&
216 cpe->ch[0].band_type[w*16 + g] < NOISE_BT &&
217 cpe->ch[1].band_type[w*16 + g] < NOISE_BT) {
218 for (i = 0; i < ics->swb_sizes[g]; i++) {
219 float L = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) * 0.5f;
220 float R = L - cpe->ch[1].coeffs[start+i];
221 cpe->ch[0].coeffs[start+i] = L;
222 cpe->ch[1].coeffs[start+i] = R;
225 start += ics->swb_sizes[g];
231 for (ch = 0; ch < chans; ch++) {
232 IndividualChannelStream *ics = &cpe->ch[ch].ics;
234 cpe->ch[ch].pulse.num_pulse = 0;
235 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
236 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
237 for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
239 maxsfb = FFMAX(maxsfb, cmaxsfb);
242 ics->max_sfb = maxsfb;
244 //adjust zero bands for window groups
245 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
246 for (g = 0; g < ics->max_sfb; g++) {
248 for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
249 if (!cpe->ch[ch].zeroes[w2*16 + g]) {
254 cpe->ch[ch].zeroes[w*16 + g] = i;
259 if (chans > 1 && cpe->common_window) {
260 IndividualChannelStream *ics0 = &cpe->ch[0].ics;
261 IndividualChannelStream *ics1 = &cpe->ch[1].ics;
263 ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
264 ics1->max_sfb = ics0->max_sfb;
265 for (w = 0; w < ics0->num_windows*16; w += 16)
266 for (i = 0; i < ics0->max_sfb; i++)
267 if (cpe->ms_mask[w+i])
269 if (msc == 0 || ics0->max_sfb == 0)
272 cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
277 * Encode scalefactor band coding type.
279 static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
283 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
284 s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
288 * Encode scalefactors.
290 static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
291 SingleChannelElement *sce)
293 int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;
294 int off_is = 0, noise_flag = 1;
297 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
298 for (i = 0; i < sce->ics.max_sfb; i++) {
299 if (!sce->zeroes[w*16 + i]) {
300 if (sce->band_type[w*16 + i] == NOISE_BT) {
301 diff = sce->sf_idx[w*16 + i] - off_pns;
302 off_pns = sce->sf_idx[w*16 + i];
303 if (noise_flag-- > 0) {
304 put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);
307 } else if (sce->band_type[w*16 + i] == INTENSITY_BT ||
308 sce->band_type[w*16 + i] == INTENSITY_BT2) {
309 diff = sce->sf_idx[w*16 + i] - off_is;
310 off_is = sce->sf_idx[w*16 + i];
312 diff = sce->sf_idx[w*16 + i] - off_sf;
313 off_sf = sce->sf_idx[w*16 + i];
315 diff += SCALE_DIFF_ZERO;
316 av_assert0(diff >= 0 && diff <= 120);
317 put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
326 static void encode_pulses(AACEncContext *s, Pulse *pulse)
330 put_bits(&s->pb, 1, !!pulse->num_pulse);
331 if (!pulse->num_pulse)
334 put_bits(&s->pb, 2, pulse->num_pulse - 1);
335 put_bits(&s->pb, 6, pulse->start);
336 for (i = 0; i < pulse->num_pulse; i++) {
337 put_bits(&s->pb, 5, pulse->pos[i]);
338 put_bits(&s->pb, 4, pulse->amp[i]);
343 * Encode spectral coefficients processed by psychoacoustic model.
345 static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
349 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
351 for (i = 0; i < sce->ics.max_sfb; i++) {
352 if (sce->zeroes[w*16 + i]) {
353 start += sce->ics.swb_sizes[i];
356 for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) {
357 s->coder->quantize_and_encode_band(s, &s->pb,
358 &sce->coeffs[start + w2*128],
359 NULL, sce->ics.swb_sizes[i],
360 sce->sf_idx[w*16 + i],
361 sce->band_type[w*16 + i],
363 sce->ics.window_clipping[w]);
365 start += sce->ics.swb_sizes[i];
371 * Downscale spectral coefficients for near-clipping windows to avoid artifacts
373 static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)
377 if (sce->ics.clip_avoidance_factor < 1.0f) {
378 for (w = 0; w < sce->ics.num_windows; w++) {
380 for (i = 0; i < sce->ics.max_sfb; i++) {
381 float *swb_coeffs = &sce->coeffs[start + w*128];
382 for (j = 0; j < sce->ics.swb_sizes[i]; j++)
383 swb_coeffs[j] *= sce->ics.clip_avoidance_factor;
384 start += sce->ics.swb_sizes[i];
391 * Encode one channel of audio data.
393 static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
394 SingleChannelElement *sce,
397 put_bits(&s->pb, 8, sce->sf_idx[0]);
398 if (!common_window) {
399 put_ics_info(s, &sce->ics);
400 if (s->coder->encode_main_pred)
401 s->coder->encode_main_pred(s, sce);
403 encode_band_info(s, sce);
404 encode_scale_factors(avctx, s, sce);
405 encode_pulses(s, &sce->pulse);
406 put_bits(&s->pb, 1, !!sce->tns.present);
407 if (s->coder->encode_tns_info)
408 s->coder->encode_tns_info(s, sce);
409 put_bits(&s->pb, 1, 0); //ssr
410 encode_spectral_coeffs(s, sce);
415 * Write some auxiliary information about the created AAC file.
417 static void put_bitstream_info(AACEncContext *s, const char *name)
419 int i, namelen, padbits;
421 namelen = strlen(name) + 2;
422 put_bits(&s->pb, 3, TYPE_FIL);
423 put_bits(&s->pb, 4, FFMIN(namelen, 15));
425 put_bits(&s->pb, 8, namelen - 14);
426 put_bits(&s->pb, 4, 0); //extension type - filler
427 padbits = -put_bits_count(&s->pb) & 7;
428 avpriv_align_put_bits(&s->pb);
429 for (i = 0; i < namelen - 2; i++)
430 put_bits(&s->pb, 8, name[i]);
431 put_bits(&s->pb, 12 - padbits, 0);
435 * Copy input samples.
436 * Channels are reordered from libavcodec's default order to AAC order.
438 static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
441 int end = 2048 + (frame ? frame->nb_samples : 0);
442 const uint8_t *channel_map = aac_chan_maps[s->channels - 1];
444 /* copy and remap input samples */
445 for (ch = 0; ch < s->channels; ch++) {
446 /* copy last 1024 samples of previous frame to the start of the current frame */
447 memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
449 /* copy new samples and zero any remaining samples */
451 memcpy(&s->planar_samples[ch][2048],
452 frame->extended_data[channel_map[ch]],
453 frame->nb_samples * sizeof(s->planar_samples[0][0]));
455 memset(&s->planar_samples[ch][end], 0,
456 (3072 - end) * sizeof(s->planar_samples[0][0]));
460 static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
461 const AVFrame *frame, int *got_packet_ptr)
463 AACEncContext *s = avctx->priv_data;
464 float **samples = s->planar_samples, *samples2, *la, *overlap;
466 SingleChannelElement *sce;
467 int i, ch, w, g, chans, tag, start_ch, ret;
468 int ms_mode = 0, is_mode = 0, tns_mode = 0, pred_mode = 0;
469 int chan_el_counter[4];
470 FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
472 if (s->last_frame == 2)
475 /* add current frame to queue */
477 if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
481 copy_input_samples(s, frame);
483 ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
485 if (!avctx->frame_number)
489 for (i = 0; i < s->chan_map[0]; i++) {
490 FFPsyWindowInfo* wi = windows + start_ch;
491 tag = s->chan_map[i+1];
492 chans = tag == TYPE_CPE ? 2 : 1;
494 for (ch = 0; ch < chans; ch++) {
495 IndividualChannelStream *ics = &cpe->ch[ch].ics;
496 int cur_channel = start_ch + ch;
497 float clip_avoidance_factor;
498 overlap = &samples[cur_channel][0];
499 samples2 = overlap + 1024;
500 la = samples2 + (448+64);
503 if (tag == TYPE_LFE) {
504 wi[ch].window_type[0] = ONLY_LONG_SEQUENCE;
505 wi[ch].window_shape = 0;
506 wi[ch].num_windows = 1;
507 wi[ch].grouping[0] = 1;
509 /* Only the lowest 12 coefficients are used in a LFE channel.
510 * The expression below results in only the bottom 8 coefficients
511 * being used for 11.025kHz to 16kHz sample rates.
513 ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
515 wi[ch] = s->psy.model->window(&s->psy, samples2, la, cur_channel,
516 ics->window_sequence[0]);
518 ics->window_sequence[1] = ics->window_sequence[0];
519 ics->window_sequence[0] = wi[ch].window_type[0];
520 ics->use_kb_window[1] = ics->use_kb_window[0];
521 ics->use_kb_window[0] = wi[ch].window_shape;
522 ics->num_windows = wi[ch].num_windows;
523 ics->swb_sizes = s->psy.bands [ics->num_windows == 8];
524 ics->num_swb = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
525 ics->swb_offset = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
526 ff_swb_offset_128 [s->samplerate_index]:
527 ff_swb_offset_1024[s->samplerate_index];
528 clip_avoidance_factor = 0.0f;
529 for (w = 0; w < ics->num_windows; w++)
530 ics->group_len[w] = wi[ch].grouping[w];
531 for (w = 0; w < ics->num_windows; w++) {
532 if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
533 ics->window_clipping[w] = 1;
534 clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
536 ics->window_clipping[w] = 0;
539 if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
540 ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
542 ics->clip_avoidance_factor = 1.0f;
545 apply_window_and_mdct(s, &cpe->ch[ch], overlap);
546 if (isnan(cpe->ch->coeffs[0])) {
547 av_log(avctx, AV_LOG_ERROR, "Input contains NaN\n");
548 return AVERROR(EINVAL);
550 avoid_clipping(s, &cpe->ch[ch]);
554 if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)
559 init_put_bits(&s->pb, avpkt->data, avpkt->size);
561 if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
562 put_bitstream_info(s, LIBAVCODEC_IDENT);
564 memset(chan_el_counter, 0, sizeof(chan_el_counter));
565 for (i = 0; i < s->chan_map[0]; i++) {
566 FFPsyWindowInfo* wi = windows + start_ch;
567 const float *coeffs[2];
568 tag = s->chan_map[i+1];
569 chans = tag == TYPE_CPE ? 2 : 1;
571 memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
572 memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
573 put_bits(&s->pb, 3, tag);
574 put_bits(&s->pb, 4, chan_el_counter[tag]++);
575 for (ch = 0; ch < chans; ch++) {
577 coeffs[ch] = sce->coeffs;
578 sce->ics.predictor_present = 0;
579 memset(&sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));
580 memset(&sce->tns, 0, sizeof(TemporalNoiseShaping));
581 for (w = 0; w < 128; w++)
582 if (sce->band_type[w] > RESERVED_BT)
583 sce->band_type[w] = 0;
585 s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
586 for (ch = 0; ch < chans; ch++) {
587 s->cur_channel = start_ch + ch;
588 s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
590 cpe->common_window = 0;
592 && wi[0].window_type[0] == wi[1].window_type[0]
593 && wi[0].window_shape == wi[1].window_shape) {
595 cpe->common_window = 1;
596 for (w = 0; w < wi[0].num_windows; w++) {
597 if (wi[0].grouping[w] != wi[1].grouping[w]) {
598 cpe->common_window = 0;
603 for (ch = 0; ch < chans; ch++) {
605 s->cur_channel = start_ch + ch;
606 if (s->options.pns && s->coder->search_for_pns)
607 s->coder->search_for_pns(s, avctx, sce);
608 if (s->options.tns && s->coder->search_for_tns)
609 s->coder->search_for_tns(s, sce);
610 if (s->options.tns && s->coder->apply_tns_filt)
611 s->coder->apply_tns_filt(sce);
612 if (sce->tns.present)
615 s->cur_channel = start_ch;
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);
626 if (s->options.intensity_stereo && s->coder->search_for_is) {
627 s->coder->search_for_is(s, avctx, cpe);
628 if (cpe->is_mode) is_mode = 1;
630 if (s->coder->set_special_band_scalefactors)
631 for (ch = 0; ch < chans; ch++)
632 s->coder->set_special_band_scalefactors(s, &cpe->ch[ch]);
633 adjust_frame_information(cpe, chans);
634 for (ch = 0; ch < chans; ch++) {
636 s->cur_channel = start_ch + ch;
637 if (s->options.pred && s->coder->search_for_pred)
638 s->coder->search_for_pred(s, sce);
639 if (cpe->ch[ch].ics.predictor_present) pred_mode = 1;
641 if (s->options.pred && s->coder->adjust_common_prediction)
642 s->coder->adjust_common_prediction(s, cpe);
643 for (ch = 0; ch < chans; ch++) {
645 s->cur_channel = start_ch + ch;
646 if (s->options.pred && s->coder->apply_main_pred)
647 s->coder->apply_main_pred(s, sce);
649 s->cur_channel = start_ch;
651 put_bits(&s->pb, 1, cpe->common_window);
652 if (cpe->common_window) {
653 put_ics_info(s, &cpe->ch[0].ics);
654 if (s->coder->encode_main_pred)
655 s->coder->encode_main_pred(s, &cpe->ch[0]);
656 encode_ms_info(&s->pb, cpe);
657 if (cpe->ms_mode) ms_mode = 1;
660 for (ch = 0; ch < chans; ch++) {
661 s->cur_channel = start_ch + ch;
662 encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
667 frame_bits = put_bits_count(&s->pb);
668 if (frame_bits <= 6144 * s->channels - 3) {
669 s->psy.bitres.bits = frame_bits / s->channels;
672 if (is_mode || ms_mode || tns_mode || pred_mode) {
673 for (i = 0; i < s->chan_map[0]; i++) {
674 // Must restore coeffs
675 chans = tag == TYPE_CPE ? 2 : 1;
677 for (ch = 0; ch < chans; ch++)
678 memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
682 s->lambda *= avctx->bit_rate * 1024.0f / avctx->sample_rate / frame_bits;
686 put_bits(&s->pb, 3, TYPE_END);
687 flush_put_bits(&s->pb);
688 avctx->frame_bits = put_bits_count(&s->pb);
690 // rate control stuff
691 if (!(avctx->flags & AV_CODEC_FLAG_QSCALE)) {
692 float ratio = avctx->bit_rate * 1024.0f / avctx->sample_rate / avctx->frame_bits;
694 s->lambda = FFMIN(s->lambda, 65536.f);
700 ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
703 avpkt->size = put_bits_count(&s->pb) >> 3;
708 static av_cold int aac_encode_end(AVCodecContext *avctx)
710 AACEncContext *s = avctx->priv_data;
712 ff_mdct_end(&s->mdct1024);
713 ff_mdct_end(&s->mdct128);
717 ff_psy_preprocess_end(s->psypp);
718 av_freep(&s->buffer.samples);
721 ff_af_queue_close(&s->afq);
725 static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
729 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
731 return AVERROR(ENOMEM);
734 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
735 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
736 ff_init_ff_sine_windows(10);
737 ff_init_ff_sine_windows(7);
739 if ((ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) < 0)
741 if ((ret = ff_mdct_init(&s->mdct128, 8, 0, 32768.0)) < 0)
747 static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
750 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->buffer.samples, s->channels, 3 * 1024 * sizeof(s->buffer.samples[0]), alloc_fail);
751 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->cpe, s->chan_map[0], sizeof(ChannelElement), alloc_fail);
752 FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + AV_INPUT_BUFFER_PADDING_SIZE, alloc_fail);
754 for(ch = 0; ch < s->channels; ch++)
755 s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
759 return AVERROR(ENOMEM);
762 static av_cold int aac_encode_init(AVCodecContext *avctx)
764 AACEncContext *s = avctx->priv_data;
766 const uint8_t *sizes[2];
767 uint8_t grouping[AAC_MAX_CHANNELS];
770 avctx->frame_size = 1024;
772 for (i = 0; i < 16; i++)
773 if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
776 s->channels = avctx->channels;
778 ERROR_IF(i == 16 || i >= ff_aac_swb_size_1024_len || i >= ff_aac_swb_size_128_len,
779 "Unsupported sample rate %d\n", avctx->sample_rate);
780 ERROR_IF(s->channels > AAC_MAX_CHANNELS,
781 "Unsupported number of channels: %d\n", s->channels);
782 WARN_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
783 "Too many bits per frame requested, clamping to max\n");
784 if (avctx->profile == FF_PROFILE_AAC_MAIN) {
786 } else if ((avctx->profile == FF_PROFILE_AAC_LOW ||
787 avctx->profile == FF_PROFILE_UNKNOWN) && s->options.pred) {
788 s->profile = 0; /* Main */
789 WARN_IF(1, "Prediction requested, changing profile to AAC-Main\n");
790 } else if (avctx->profile == FF_PROFILE_AAC_LOW ||
791 avctx->profile == FF_PROFILE_UNKNOWN) {
792 s->profile = 1; /* Low */
794 ERROR_IF(1, "Unsupported profile %d\n", avctx->profile);
797 avctx->bit_rate = (int)FFMIN(
798 6144 * s->channels / 1024.0 * avctx->sample_rate,
801 s->samplerate_index = i;
803 s->chan_map = aac_chan_configs[s->channels-1];
805 if ((ret = dsp_init(avctx, s)) < 0)
808 if ((ret = alloc_buffers(avctx, s)) < 0)
811 avctx->extradata_size = 5;
812 put_audio_specific_config(avctx);
814 sizes[0] = ff_aac_swb_size_1024[i];
815 sizes[1] = ff_aac_swb_size_128[i];
816 lengths[0] = ff_aac_num_swb_1024[i];
817 lengths[1] = ff_aac_num_swb_128[i];
818 for (i = 0; i < s->chan_map[0]; i++)
819 grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
820 if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
821 s->chan_map[0], grouping)) < 0)
823 s->psypp = ff_psy_preprocess_init(avctx);
824 s->coder = &ff_aac_coders[s->options.aac_coder];
825 ff_lpc_init(&s->lpc, avctx->frame_size, MAX_LPC_ORDER, FF_LPC_TYPE_LEVINSON);
828 ff_aac_coder_init_mips(s);
830 s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
834 avctx->initial_padding = 1024;
835 ff_af_queue_init(avctx, &s->afq);
839 aac_encode_end(avctx);
843 #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
844 static const AVOption aacenc_options[] = {
845 {"stereo_mode", "Stereo coding method", offsetof(AACEncContext, options.stereo_mode), AV_OPT_TYPE_INT, {.i64 = 0}, -1, 1, AACENC_FLAGS, "stereo_mode"},
846 {"auto", "Selected by the Encoder", 0, AV_OPT_TYPE_CONST, {.i64 = -1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
847 {"ms_off", "Disable Mid/Side coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
848 {"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"},
849 {"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"},
850 {"faac", "FAAC-inspired method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAAC}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
851 {"anmr", "ANMR method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
852 {"twoloop", "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
853 {"fast", "Constant quantizer", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
854 {"aac_pns", "Perceptual Noise Substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, AACENC_FLAGS, "aac_pns"},
855 {"disable", "Disable perceptual noise substitution", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pns"},
856 {"enable", "Enable perceptual noise substitution", 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pns"},
857 {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, AACENC_FLAGS, "intensity_stereo"},
858 {"disable", "Disable intensity stereo coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0}, INT_MIN, INT_MAX, AACENC_FLAGS, "intensity_stereo"},
859 {"enable", "Enable intensity stereo coding", 0, AV_OPT_TYPE_CONST, {.i64 = 1}, INT_MIN, INT_MAX, AACENC_FLAGS, "intensity_stereo"},
860 {"aac_tns", "Temporal noise shaping", offsetof(AACEncContext, options.tns), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, AACENC_FLAGS, "aac_tns"},
861 {"disable", "Disable temporal noise shaping", 0, AV_OPT_TYPE_CONST, {.i64 = 0}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_tns"},
862 {"enable", "Enable temporal noise shaping", 0, AV_OPT_TYPE_CONST, {.i64 = 1}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_tns"},
863 {"aac_pred", "AAC-Main prediction", offsetof(AACEncContext, options.pred), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, AACENC_FLAGS, "aac_pred"},
864 {"disable", "Disable AAC-Main prediction", 0, AV_OPT_TYPE_CONST, {.i64 = 0}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pred"},
865 {"enable", "Enable AAC-Main prediction", 0, AV_OPT_TYPE_CONST, {.i64 = 1}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pred"},
869 static const AVClass aacenc_class = {
871 av_default_item_name,
873 LIBAVUTIL_VERSION_INT,
876 AVCodec ff_aac_encoder = {
878 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
879 .type = AVMEDIA_TYPE_AUDIO,
880 .id = AV_CODEC_ID_AAC,
881 .priv_data_size = sizeof(AACEncContext),
882 .init = aac_encode_init,
883 .encode2 = aac_encode_frame,
884 .close = aac_encode_end,
885 .supported_samplerates = mpeg4audio_sample_rates,
886 .capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY |
887 AV_CODEC_CAP_EXPERIMENTAL,
888 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
889 AV_SAMPLE_FMT_NONE },
890 .priv_class = &aacenc_class,