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"
45 #include "aacenctab.h"
49 #define ERROR_IF(cond, ...) \
51 av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \
52 return AVERROR(EINVAL); \
55 #define WARN_IF(cond, ...) \
57 av_log(avctx, AV_LOG_WARNING, __VA_ARGS__); \
61 * Make AAC audio config object.
62 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
64 static void put_audio_specific_config(AVCodecContext *avctx)
67 AACEncContext *s = avctx->priv_data;
69 init_put_bits(&pb, avctx->extradata, avctx->extradata_size);
70 put_bits(&pb, 5, 2); //object type - AAC-LC
71 put_bits(&pb, 4, s->samplerate_index); //sample rate index
72 put_bits(&pb, 4, s->channels);
74 put_bits(&pb, 1, 0); //frame length - 1024 samples
75 put_bits(&pb, 1, 0); //does not depend on core coder
76 put_bits(&pb, 1, 0); //is not extension
78 //Explicitly Mark SBR absent
79 put_bits(&pb, 11, 0x2b7); //sync extension
80 put_bits(&pb, 5, AOT_SBR);
85 #define WINDOW_FUNC(type) \
86 static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \
87 SingleChannelElement *sce, \
90 WINDOW_FUNC(only_long)
92 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
93 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
94 float *out = sce->ret_buf;
96 fdsp->vector_fmul (out, audio, lwindow, 1024);
97 fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
100 WINDOW_FUNC(long_start)
102 const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
103 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
104 float *out = sce->ret_buf;
106 fdsp->vector_fmul(out, audio, lwindow, 1024);
107 memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
108 fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
109 memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
112 WINDOW_FUNC(long_stop)
114 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
115 const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
116 float *out = sce->ret_buf;
118 memset(out, 0, sizeof(out[0]) * 448);
119 fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
120 memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
121 fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
124 WINDOW_FUNC(eight_short)
126 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
127 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
128 const float *in = audio + 448;
129 float *out = sce->ret_buf;
132 for (w = 0; w < 8; w++) {
133 fdsp->vector_fmul (out, in, w ? pwindow : swindow, 128);
136 fdsp->vector_fmul_reverse(out, in, swindow, 128);
141 static void (*const apply_window[4])(AVFloatDSPContext *fdsp,
142 SingleChannelElement *sce,
143 const float *audio) = {
144 [ONLY_LONG_SEQUENCE] = apply_only_long_window,
145 [LONG_START_SEQUENCE] = apply_long_start_window,
146 [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
147 [LONG_STOP_SEQUENCE] = apply_long_stop_window
150 static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
154 float *output = sce->ret_buf;
156 apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);
158 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
159 s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
161 for (i = 0; i < 1024; i += 128)
162 s->mdct128.mdct_calc(&s->mdct128, sce->coeffs + i, output + i*2);
163 memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
164 memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));
168 * Encode ics_info element.
169 * @see Table 4.6 (syntax of ics_info)
171 static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
175 put_bits(&s->pb, 1, 0); // ics_reserved bit
176 put_bits(&s->pb, 2, info->window_sequence[0]);
177 put_bits(&s->pb, 1, info->use_kb_window[0]);
178 if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
179 put_bits(&s->pb, 6, info->max_sfb);
180 put_bits(&s->pb, 1, 0); // no prediction
182 put_bits(&s->pb, 4, info->max_sfb);
183 for (w = 1; w < 8; w++)
184 put_bits(&s->pb, 1, !info->group_len[w]);
190 * @see 4.6.8.1 "Joint Coding - M/S Stereo"
192 static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
196 put_bits(pb, 2, cpe->ms_mode);
197 if (cpe->ms_mode == 1)
198 for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
199 for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
200 put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
204 * Produce integer coefficients from scalefactors provided by the model.
206 static void adjust_frame_information(ChannelElement *cpe, int chans)
210 IndividualChannelStream *ics;
212 if (cpe->common_window) {
213 ics = &cpe->ch[0].ics;
214 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
215 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
216 int start = (w+w2) * 128;
217 for (g = 0; g < ics->num_swb; g++) {
218 //apply Intensity stereo coeffs transformation
219 if (cpe->is_mask[w*16 + g]) {
220 int p = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
221 float scale = cpe->ch[0].is_ener[w*16+g];
222 for (i = 0; i < ics->swb_sizes[g]; i++) {
223 cpe->ch[0].coeffs[start+i] = (cpe->ch[0].pcoeffs[start+i] + p*cpe->ch[1].pcoeffs[start+i]) * scale;
224 cpe->ch[1].coeffs[start+i] = 0.0f;
226 } else if (cpe->ms_mask[w*16 + g] &&
227 cpe->ch[0].band_type[w*16 + g] < NOISE_BT &&
228 cpe->ch[1].band_type[w*16 + g] < NOISE_BT) {
229 for (i = 0; i < ics->swb_sizes[g]; i++) {
230 cpe->ch[0].coeffs[start+i] = (cpe->ch[0].pcoeffs[start+i] + cpe->ch[1].pcoeffs[start+i]) * 0.5f;
231 cpe->ch[1].coeffs[start+i] = cpe->ch[0].coeffs[start+i] - cpe->ch[1].pcoeffs[start+i];
234 start += ics->swb_sizes[g];
240 for (ch = 0; ch < chans; ch++) {
241 IndividualChannelStream *ics = &cpe->ch[ch].ics;
243 cpe->ch[ch].pulse.num_pulse = 0;
244 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
245 for (w2 = 0; w2 < ics->group_len[w]; w2++) {
246 for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
248 maxsfb = FFMAX(maxsfb, cmaxsfb);
251 ics->max_sfb = maxsfb;
253 //adjust zero bands for window groups
254 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
255 for (g = 0; g < ics->max_sfb; g++) {
257 for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
258 if (!cpe->ch[ch].zeroes[w2*16 + g]) {
263 cpe->ch[ch].zeroes[w*16 + g] = i;
268 if (chans > 1 && cpe->common_window) {
269 IndividualChannelStream *ics0 = &cpe->ch[0].ics;
270 IndividualChannelStream *ics1 = &cpe->ch[1].ics;
272 ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
273 ics1->max_sfb = ics0->max_sfb;
274 for (w = 0; w < ics0->num_windows*16; w += 16)
275 for (i = 0; i < ics0->max_sfb; i++)
276 if (cpe->ms_mask[w+i])
278 if (msc == 0 || ics0->max_sfb == 0)
281 cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
286 * Encode scalefactor band coding type.
288 static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
292 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
293 s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
297 * Encode scalefactors.
299 static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
300 SingleChannelElement *sce)
302 int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;
303 int off_is = 0, noise_flag = 1;
306 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
307 for (i = 0; i < sce->ics.max_sfb; i++) {
308 if (!sce->zeroes[w*16 + i]) {
309 if (sce->band_type[w*16 + i] == NOISE_BT) {
310 diff = sce->sf_idx[w*16 + i] - off_pns;
311 off_pns = sce->sf_idx[w*16 + i];
312 if (noise_flag-- > 0) {
313 put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);
316 } else if (sce->band_type[w*16 + i] == INTENSITY_BT ||
317 sce->band_type[w*16 + i] == INTENSITY_BT2) {
318 diff = sce->sf_idx[w*16 + i] - off_is;
319 off_is = sce->sf_idx[w*16 + i];
321 diff = sce->sf_idx[w*16 + i] - off_sf;
322 off_sf = sce->sf_idx[w*16 + i];
324 diff += SCALE_DIFF_ZERO;
325 av_assert0(diff >= 0 && diff <= 120);
326 put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
335 static void encode_pulses(AACEncContext *s, Pulse *pulse)
339 put_bits(&s->pb, 1, !!pulse->num_pulse);
340 if (!pulse->num_pulse)
343 put_bits(&s->pb, 2, pulse->num_pulse - 1);
344 put_bits(&s->pb, 6, pulse->start);
345 for (i = 0; i < pulse->num_pulse; i++) {
346 put_bits(&s->pb, 5, pulse->pos[i]);
347 put_bits(&s->pb, 4, pulse->amp[i]);
352 * Encode spectral coefficients processed by psychoacoustic model.
354 static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
358 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
360 for (i = 0; i < sce->ics.max_sfb; i++) {
361 if (sce->zeroes[w*16 + i]) {
362 start += sce->ics.swb_sizes[i];
365 for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++)
366 s->coder->quantize_and_encode_band(s, &s->pb, sce->coeffs + start + w2*128,
367 sce->ics.swb_sizes[i],
368 sce->sf_idx[w*16 + i],
369 sce->band_type[w*16 + i],
370 s->lambda, sce->ics.window_clipping[w]);
371 start += sce->ics.swb_sizes[i];
377 * Downscale spectral coefficients for near-clipping windows to avoid artifacts
379 static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)
383 if (sce->ics.clip_avoidance_factor < 1.0f) {
384 for (w = 0; w < sce->ics.num_windows; w++) {
386 for (i = 0; i < sce->ics.max_sfb; i++) {
387 float *swb_coeffs = sce->coeffs + start + w*128;
388 for (j = 0; j < sce->ics.swb_sizes[i]; j++)
389 swb_coeffs[j] *= sce->ics.clip_avoidance_factor;
390 start += sce->ics.swb_sizes[i];
397 * Encode one channel of audio data.
399 static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
400 SingleChannelElement *sce,
403 put_bits(&s->pb, 8, sce->sf_idx[0]);
405 put_ics_info(s, &sce->ics);
406 encode_band_info(s, sce);
407 encode_scale_factors(avctx, s, sce);
408 encode_pulses(s, &sce->pulse);
409 put_bits(&s->pb, 1, 0); //tns
410 put_bits(&s->pb, 1, 0); //ssr
411 encode_spectral_coeffs(s, sce);
416 * Write some auxiliary information about the created AAC file.
418 static void put_bitstream_info(AACEncContext *s, const char *name)
420 int i, namelen, padbits;
422 namelen = strlen(name) + 2;
423 put_bits(&s->pb, 3, TYPE_FIL);
424 put_bits(&s->pb, 4, FFMIN(namelen, 15));
426 put_bits(&s->pb, 8, namelen - 14);
427 put_bits(&s->pb, 4, 0); //extension type - filler
428 padbits = -put_bits_count(&s->pb) & 7;
429 avpriv_align_put_bits(&s->pb);
430 for (i = 0; i < namelen - 2; i++)
431 put_bits(&s->pb, 8, name[i]);
432 put_bits(&s->pb, 12 - padbits, 0);
436 * Copy input samples.
437 * Channels are reordered from libavcodec's default order to AAC order.
439 static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
442 int end = 2048 + (frame ? frame->nb_samples : 0);
443 const uint8_t *channel_map = aac_chan_maps[s->channels - 1];
445 /* copy and remap input samples */
446 for (ch = 0; ch < s->channels; ch++) {
447 /* copy last 1024 samples of previous frame to the start of the current frame */
448 memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
450 /* copy new samples and zero any remaining samples */
452 memcpy(&s->planar_samples[ch][2048],
453 frame->extended_data[channel_map[ch]],
454 frame->nb_samples * sizeof(s->planar_samples[0][0]));
456 memset(&s->planar_samples[ch][end], 0,
457 (3072 - end) * sizeof(s->planar_samples[0][0]));
461 static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
462 const AVFrame *frame, int *got_packet_ptr)
464 AACEncContext *s = avctx->priv_data;
465 float **samples = s->planar_samples, *samples2, *la, *overlap;
467 int i, ch, w, g, chans, tag, start_ch, ret, ms_mode = 0, is_mode = 0;
468 int chan_el_counter[4];
469 FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
471 if (s->last_frame == 2)
474 /* add current frame to queue */
476 if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
480 copy_input_samples(s, frame);
482 ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
484 if (!avctx->frame_number)
488 for (i = 0; i < s->chan_map[0]; i++) {
489 FFPsyWindowInfo* wi = windows + start_ch;
490 tag = s->chan_map[i+1];
491 chans = tag == TYPE_CPE ? 2 : 1;
493 for (ch = 0; ch < chans; ch++) {
494 IndividualChannelStream *ics = &cpe->ch[ch].ics;
495 int cur_channel = start_ch + ch;
496 float clip_avoidance_factor;
497 overlap = &samples[cur_channel][0];
498 samples2 = overlap + 1024;
499 la = samples2 + (448+64);
502 if (tag == TYPE_LFE) {
503 wi[ch].window_type[0] = ONLY_LONG_SEQUENCE;
504 wi[ch].window_shape = 0;
505 wi[ch].num_windows = 1;
506 wi[ch].grouping[0] = 1;
508 /* Only the lowest 12 coefficients are used in a LFE channel.
509 * The expression below results in only the bottom 8 coefficients
510 * being used for 11.025kHz to 16kHz sample rates.
512 ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
514 wi[ch] = s->psy.model->window(&s->psy, samples2, la, cur_channel,
515 ics->window_sequence[0]);
517 ics->window_sequence[1] = ics->window_sequence[0];
518 ics->window_sequence[0] = wi[ch].window_type[0];
519 ics->use_kb_window[1] = ics->use_kb_window[0];
520 ics->use_kb_window[0] = wi[ch].window_shape;
521 ics->num_windows = wi[ch].num_windows;
522 ics->swb_sizes = s->psy.bands [ics->num_windows == 8];
523 ics->num_swb = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
524 clip_avoidance_factor = 0.0f;
525 for (w = 0; w < ics->num_windows; w++)
526 ics->group_len[w] = wi[ch].grouping[w];
527 for (w = 0; w < ics->num_windows; w++) {
528 if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
529 ics->window_clipping[w] = 1;
530 clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
532 ics->window_clipping[w] = 0;
535 if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
536 ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
538 ics->clip_avoidance_factor = 1.0f;
541 apply_window_and_mdct(s, &cpe->ch[ch], overlap);
542 if (isnan(cpe->ch->coeffs[0])) {
543 av_log(avctx, AV_LOG_ERROR, "Input contains NaN\n");
544 return AVERROR(EINVAL);
546 avoid_clipping(s, &cpe->ch[ch]);
550 if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)
555 init_put_bits(&s->pb, avpkt->data, avpkt->size);
557 if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
558 put_bitstream_info(s, LIBAVCODEC_IDENT);
560 memset(chan_el_counter, 0, sizeof(chan_el_counter));
561 for (i = 0; i < s->chan_map[0]; i++) {
562 FFPsyWindowInfo* wi = windows + start_ch;
563 const float *coeffs[2];
564 tag = s->chan_map[i+1];
565 chans = tag == TYPE_CPE ? 2 : 1;
567 memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
568 memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
569 put_bits(&s->pb, 3, tag);
570 put_bits(&s->pb, 4, chan_el_counter[tag]++);
571 for (ch = 0; ch < chans; ch++)
572 coeffs[ch] = cpe->ch[ch].coeffs;
573 s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
574 for (ch = 0; ch < chans; ch++) {
575 s->cur_channel = start_ch + ch;
576 s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
578 cpe->common_window = 0;
580 && wi[0].window_type[0] == wi[1].window_type[0]
581 && wi[0].window_shape == wi[1].window_shape) {
583 cpe->common_window = 1;
584 for (w = 0; w < wi[0].num_windows; w++) {
585 if (wi[0].grouping[w] != wi[1].grouping[w]) {
586 cpe->common_window = 0;
591 if (s->options.pns && s->coder->search_for_pns) {
592 for (ch = 0; ch < chans; ch++) {
593 s->cur_channel = start_ch + ch;
594 s->coder->search_for_pns(s, avctx, &cpe->ch[ch]);
597 s->cur_channel = start_ch;
598 if (s->options.stereo_mode && cpe->common_window) {
599 if (s->options.stereo_mode > 0) {
600 IndividualChannelStream *ics = &cpe->ch[0].ics;
601 for (w = 0; w < ics->num_windows; w += ics->group_len[w])
602 for (g = 0; g < ics->num_swb; g++)
603 cpe->ms_mask[w*16+g] = 1;
604 } else if (s->coder->search_for_ms) {
605 s->coder->search_for_ms(s, cpe);
608 if (chans > 1 && s->options.intensity_stereo && s->coder->search_for_is) {
609 s->coder->search_for_is(s, avctx, cpe);
610 if (cpe->is_mode) is_mode = 1;
612 if (s->coder->set_special_band_scalefactors)
613 for (ch = 0; ch < chans; ch++)
614 s->coder->set_special_band_scalefactors(s, &cpe->ch[ch]);
615 adjust_frame_information(cpe, chans);
617 put_bits(&s->pb, 1, cpe->common_window);
618 if (cpe->common_window) {
619 put_ics_info(s, &cpe->ch[0].ics);
620 encode_ms_info(&s->pb, cpe);
621 if (cpe->ms_mode) ms_mode = 1;
624 for (ch = 0; ch < chans; ch++) {
625 s->cur_channel = start_ch + ch;
626 encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
631 frame_bits = put_bits_count(&s->pb);
632 if (frame_bits <= 6144 * s->channels - 3) {
633 s->psy.bitres.bits = frame_bits / s->channels;
636 if (is_mode || ms_mode) {
637 for (i = 0; i < s->chan_map[0]; i++) {
638 // Must restore coeffs
639 chans = tag == TYPE_CPE ? 2 : 1;
641 for (ch = 0; ch < chans; ch++)
642 memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
646 s->lambda *= avctx->bit_rate * 1024.0f / avctx->sample_rate / frame_bits;
650 put_bits(&s->pb, 3, TYPE_END);
651 flush_put_bits(&s->pb);
652 avctx->frame_bits = put_bits_count(&s->pb);
654 // rate control stuff
655 if (!(avctx->flags & AV_CODEC_FLAG_QSCALE)) {
656 float ratio = avctx->bit_rate * 1024.0f / avctx->sample_rate / avctx->frame_bits;
658 s->lambda = FFMIN(s->lambda, 65536.f);
664 ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
667 avpkt->size = put_bits_count(&s->pb) >> 3;
672 static av_cold int aac_encode_end(AVCodecContext *avctx)
674 AACEncContext *s = avctx->priv_data;
676 ff_mdct_end(&s->mdct1024);
677 ff_mdct_end(&s->mdct128);
680 ff_psy_preprocess_end(s->psypp);
681 av_freep(&s->buffer.samples);
684 ff_af_queue_close(&s->afq);
688 static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
692 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
694 return AVERROR(ENOMEM);
697 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
698 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
699 ff_init_ff_sine_windows(10);
700 ff_init_ff_sine_windows(7);
702 if ((ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) < 0)
704 if ((ret = ff_mdct_init(&s->mdct128, 8, 0, 32768.0)) < 0)
710 static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
713 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->buffer.samples, s->channels, 3 * 1024 * sizeof(s->buffer.samples[0]), alloc_fail);
714 FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->cpe, s->chan_map[0], sizeof(ChannelElement), alloc_fail);
715 FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + AV_INPUT_BUFFER_PADDING_SIZE, alloc_fail);
717 for(ch = 0; ch < s->channels; ch++)
718 s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
722 return AVERROR(ENOMEM);
725 static av_cold int aac_encode_init(AVCodecContext *avctx)
727 AACEncContext *s = avctx->priv_data;
729 const uint8_t *sizes[2];
730 uint8_t grouping[AAC_MAX_CHANNELS];
733 avctx->frame_size = 1024;
735 for (i = 0; i < 16; i++)
736 if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
739 s->channels = avctx->channels;
741 ERROR_IF(i == 16 || i >= swb_size_1024_len || i >= swb_size_128_len,
742 "Unsupported sample rate %d\n", avctx->sample_rate);
743 ERROR_IF(s->channels > AAC_MAX_CHANNELS,
744 "Unsupported number of channels: %d\n", s->channels);
745 ERROR_IF(avctx->profile != FF_PROFILE_UNKNOWN && avctx->profile != FF_PROFILE_AAC_LOW,
746 "Unsupported profile %d\n", avctx->profile);
747 WARN_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
748 "Too many bits per frame requested, clamping to max\n");
750 avctx->bit_rate = (int)FFMIN(
751 6144 * s->channels / 1024.0 * avctx->sample_rate,
754 s->samplerate_index = i;
756 s->chan_map = aac_chan_configs[s->channels-1];
758 if ((ret = dsp_init(avctx, s)) < 0)
761 if ((ret = alloc_buffers(avctx, s)) < 0)
764 avctx->extradata_size = 5;
765 put_audio_specific_config(avctx);
767 sizes[0] = swb_size_1024[i];
768 sizes[1] = swb_size_128[i];
769 lengths[0] = ff_aac_num_swb_1024[i];
770 lengths[1] = ff_aac_num_swb_128[i];
771 for (i = 0; i < s->chan_map[0]; i++)
772 grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
773 if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
774 s->chan_map[0], grouping)) < 0)
776 s->psypp = ff_psy_preprocess_init(avctx);
777 s->coder = &ff_aac_coders[s->options.aac_coder];
780 ff_aac_coder_init_mips(s);
782 s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
786 avctx->initial_padding = 1024;
787 ff_af_queue_init(avctx, &s->afq);
791 aac_encode_end(avctx);
795 #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
796 static const AVOption aacenc_options[] = {
797 {"stereo_mode", "Stereo coding method", offsetof(AACEncContext, options.stereo_mode), AV_OPT_TYPE_INT, {.i64 = 0}, -1, 1, AACENC_FLAGS, "stereo_mode"},
798 {"auto", "Selected by the Encoder", 0, AV_OPT_TYPE_CONST, {.i64 = -1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
799 {"ms_off", "Disable Mid/Side coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
800 {"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"},
801 {"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"},
802 {"faac", "FAAC-inspired method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAAC}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
803 {"anmr", "ANMR method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
804 {"twoloop", "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
805 {"fast", "Constant quantizer", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
806 {"aac_pns", "Perceptual Noise Substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, AACENC_FLAGS, "aac_pns"},
807 {"disable", "Disable perceptual noise substitution", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pns"},
808 {"enable", "Enable perceptual noise substitution", 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_pns"},
809 {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, AACENC_FLAGS, "intensity_stereo"},
810 {"disable", "Disable intensity stereo coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0}, INT_MIN, INT_MAX, AACENC_FLAGS, "intensity_stereo"},
811 {"enable", "Enable intensity stereo coding", 0, AV_OPT_TYPE_CONST, {.i64 = 1}, INT_MIN, INT_MAX, AACENC_FLAGS, "intensity_stereo"},
815 static const AVClass aacenc_class = {
817 av_default_item_name,
819 LIBAVUTIL_VERSION_INT,
822 AVCodec ff_aac_encoder = {
824 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
825 .type = AVMEDIA_TYPE_AUDIO,
826 .id = AV_CODEC_ID_AAC,
827 .priv_data_size = sizeof(AACEncContext),
828 .init = aac_encode_init,
829 .encode2 = aac_encode_frame,
830 .close = aac_encode_end,
831 .supported_samplerates = mpeg4audio_sample_rates,
832 .capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY |
833 AV_CODEC_CAP_EXPERIMENTAL,
834 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
835 AV_SAMPLE_FMT_NONE },
836 .priv_class = &aacenc_class,