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"
39 #include "mpeg4audio.h"
49 #define AAC_MAX_CHANNELS 6
51 #define ERROR_IF(cond, ...) \
53 av_log(avctx, AV_LOG_ERROR, __VA_ARGS__); \
54 return AVERROR(EINVAL); \
57 float ff_aac_pow34sf_tab[428];
59 static const uint8_t swb_size_1024_96[] = {
60 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8,
61 12, 12, 12, 12, 12, 16, 16, 24, 28, 36, 44,
62 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64
65 static const uint8_t swb_size_1024_64[] = {
66 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8,
67 12, 12, 12, 16, 16, 16, 20, 24, 24, 28, 36,
68 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40
71 static const uint8_t swb_size_1024_48[] = {
72 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8,
73 12, 12, 12, 12, 16, 16, 20, 20, 24, 24, 28, 28,
74 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32,
78 static const uint8_t swb_size_1024_32[] = {
79 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8,
80 12, 12, 12, 12, 16, 16, 20, 20, 24, 24, 28, 28,
81 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32
84 static const uint8_t swb_size_1024_24[] = {
85 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
86 12, 12, 12, 12, 16, 16, 16, 20, 20, 24, 24, 28, 28,
87 32, 36, 36, 40, 44, 48, 52, 52, 64, 64, 64, 64, 64
90 static const uint8_t swb_size_1024_16[] = {
91 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
92 12, 12, 12, 12, 12, 12, 12, 12, 12, 16, 16, 16, 16, 20, 20, 20, 24, 24, 28, 28,
93 32, 36, 40, 40, 44, 48, 52, 56, 60, 64, 64, 64
96 static const uint8_t swb_size_1024_8[] = {
97 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12,
98 16, 16, 16, 16, 16, 16, 16, 20, 20, 20, 20, 24, 24, 24, 28, 28,
99 32, 36, 36, 40, 44, 48, 52, 56, 60, 64, 80
102 static const uint8_t *swb_size_1024[] = {
103 swb_size_1024_96, swb_size_1024_96, swb_size_1024_64,
104 swb_size_1024_48, swb_size_1024_48, swb_size_1024_32,
105 swb_size_1024_24, swb_size_1024_24, swb_size_1024_16,
106 swb_size_1024_16, swb_size_1024_16, swb_size_1024_8
109 static const uint8_t swb_size_128_96[] = {
110 4, 4, 4, 4, 4, 4, 8, 8, 8, 16, 28, 36
113 static const uint8_t swb_size_128_48[] = {
114 4, 4, 4, 4, 4, 8, 8, 8, 12, 12, 12, 16, 16, 16
117 static const uint8_t swb_size_128_24[] = {
118 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 20
121 static const uint8_t swb_size_128_16[] = {
122 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 12, 12, 16, 20, 20
125 static const uint8_t swb_size_128_8[] = {
126 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 12, 16, 20, 20
129 static const uint8_t *swb_size_128[] = {
130 /* the last entry on the following row is swb_size_128_64 but is a
131 duplicate of swb_size_128_96 */
132 swb_size_128_96, swb_size_128_96, swb_size_128_96,
133 swb_size_128_48, swb_size_128_48, swb_size_128_48,
134 swb_size_128_24, swb_size_128_24, swb_size_128_16,
135 swb_size_128_16, swb_size_128_16, swb_size_128_8
138 /** default channel configurations */
139 static const uint8_t aac_chan_configs[6][5] = {
140 {1, TYPE_SCE}, // 1 channel - single channel element
141 {1, TYPE_CPE}, // 2 channels - channel pair
142 {2, TYPE_SCE, TYPE_CPE}, // 3 channels - center + stereo
143 {3, TYPE_SCE, TYPE_CPE, TYPE_SCE}, // 4 channels - front center + stereo + back center
144 {3, TYPE_SCE, TYPE_CPE, TYPE_CPE}, // 5 channels - front center + stereo + back stereo
145 {4, TYPE_SCE, TYPE_CPE, TYPE_CPE, TYPE_LFE}, // 6 channels - front center + stereo + back stereo + LFE
149 * Table to remap channels from libavcodec's default order to AAC order.
151 static const uint8_t aac_chan_maps[AAC_MAX_CHANNELS][AAC_MAX_CHANNELS] = {
157 { 2, 0, 1, 4, 5, 3 },
161 * Make AAC audio config object.
162 * @see 1.6.2.1 "Syntax - AudioSpecificConfig"
164 static void put_audio_specific_config(AVCodecContext *avctx)
167 AACEncContext *s = avctx->priv_data;
169 init_put_bits(&pb, avctx->extradata, avctx->extradata_size*8);
170 put_bits(&pb, 5, 2); //object type - AAC-LC
171 put_bits(&pb, 4, s->samplerate_index); //sample rate index
172 put_bits(&pb, 4, s->channels);
174 put_bits(&pb, 1, 0); //frame length - 1024 samples
175 put_bits(&pb, 1, 0); //does not depend on core coder
176 put_bits(&pb, 1, 0); //is not extension
178 //Explicitly Mark SBR absent
179 put_bits(&pb, 11, 0x2b7); //sync extension
180 put_bits(&pb, 5, AOT_SBR);
185 #define WINDOW_FUNC(type) \
186 static void apply_ ##type ##_window(DSPContext *dsp, AVFloatDSPContext *fdsp, \
187 SingleChannelElement *sce, \
190 WINDOW_FUNC(only_long)
192 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
193 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
194 float *out = sce->ret_buf;
196 fdsp->vector_fmul (out, audio, lwindow, 1024);
197 dsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
200 WINDOW_FUNC(long_start)
202 const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
203 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
204 float *out = sce->ret_buf;
206 fdsp->vector_fmul(out, audio, lwindow, 1024);
207 memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
208 dsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
209 memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
212 WINDOW_FUNC(long_stop)
214 const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
215 const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
216 float *out = sce->ret_buf;
218 memset(out, 0, sizeof(out[0]) * 448);
219 fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
220 memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
221 dsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
224 WINDOW_FUNC(eight_short)
226 const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
227 const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
228 const float *in = audio + 448;
229 float *out = sce->ret_buf;
232 for (w = 0; w < 8; w++) {
233 fdsp->vector_fmul (out, in, w ? pwindow : swindow, 128);
236 dsp->vector_fmul_reverse(out, in, swindow, 128);
241 static void (*const apply_window[4])(DSPContext *dsp, AVFloatDSPContext *fdsp,
242 SingleChannelElement *sce,
243 const float *audio) = {
244 [ONLY_LONG_SEQUENCE] = apply_only_long_window,
245 [LONG_START_SEQUENCE] = apply_long_start_window,
246 [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
247 [LONG_STOP_SEQUENCE] = apply_long_stop_window
250 static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
254 float *output = sce->ret_buf;
256 apply_window[sce->ics.window_sequence[0]](&s->dsp, &s->fdsp, sce, audio);
258 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
259 s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
261 for (i = 0; i < 1024; i += 128)
262 s->mdct128.mdct_calc(&s->mdct128, sce->coeffs + i, output + i*2);
263 memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
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*16; w += 16) {
316 for (g = 0; g < ics->num_swb; g++) {
318 if (cpe->common_window && !ch && cpe->ms_mask[w + g]) {
319 for (i = 0; i < ics->swb_sizes[g]; i++) {
320 cpe->ch[0].coeffs[start+i] = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) / 2.0;
321 cpe->ch[1].coeffs[start+i] = cpe->ch[0].coeffs[start+i] - cpe->ch[1].coeffs[start+i];
324 start += ics->swb_sizes[g];
326 for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w+cmaxsfb-1]; cmaxsfb--)
328 maxsfb = FFMAX(maxsfb, cmaxsfb);
330 ics->max_sfb = maxsfb;
332 //adjust zero bands for window groups
333 for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
334 for (g = 0; g < ics->max_sfb; g++) {
336 for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
337 if (!cpe->ch[ch].zeroes[w2*16 + g]) {
342 cpe->ch[ch].zeroes[w*16 + g] = i;
347 if (chans > 1 && cpe->common_window) {
348 IndividualChannelStream *ics0 = &cpe->ch[0].ics;
349 IndividualChannelStream *ics1 = &cpe->ch[1].ics;
351 ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
352 ics1->max_sfb = ics0->max_sfb;
353 for (w = 0; w < ics0->num_windows*16; w += 16)
354 for (i = 0; i < ics0->max_sfb; i++)
355 if (cpe->ms_mask[w+i])
357 if (msc == 0 || ics0->max_sfb == 0)
360 cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
365 * Encode scalefactor band coding type.
367 static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
371 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
372 s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
376 * Encode scalefactors.
378 static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
379 SingleChannelElement *sce)
381 int off = sce->sf_idx[0], diff;
384 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
385 for (i = 0; i < sce->ics.max_sfb; i++) {
386 if (!sce->zeroes[w*16 + i]) {
387 diff = sce->sf_idx[w*16 + i] - off + SCALE_DIFF_ZERO;
388 av_assert0(diff >= 0 && diff <= 120);
389 off = sce->sf_idx[w*16 + i];
390 put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
399 static void encode_pulses(AACEncContext *s, Pulse *pulse)
403 put_bits(&s->pb, 1, !!pulse->num_pulse);
404 if (!pulse->num_pulse)
407 put_bits(&s->pb, 2, pulse->num_pulse - 1);
408 put_bits(&s->pb, 6, pulse->start);
409 for (i = 0; i < pulse->num_pulse; i++) {
410 put_bits(&s->pb, 5, pulse->pos[i]);
411 put_bits(&s->pb, 4, pulse->amp[i]);
416 * Encode spectral coefficients processed by psychoacoustic model.
418 static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
422 for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
424 for (i = 0; i < sce->ics.max_sfb; i++) {
425 if (sce->zeroes[w*16 + i]) {
426 start += sce->ics.swb_sizes[i];
429 for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++)
430 s->coder->quantize_and_encode_band(s, &s->pb, sce->coeffs + start + w2*128,
431 sce->ics.swb_sizes[i],
432 sce->sf_idx[w*16 + i],
433 sce->band_type[w*16 + i],
435 start += sce->ics.swb_sizes[i];
441 * Encode one channel of audio data.
443 static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
444 SingleChannelElement *sce,
447 put_bits(&s->pb, 8, sce->sf_idx[0]);
449 put_ics_info(s, &sce->ics);
450 encode_band_info(s, sce);
451 encode_scale_factors(avctx, s, sce);
452 encode_pulses(s, &sce->pulse);
453 put_bits(&s->pb, 1, 0); //tns
454 put_bits(&s->pb, 1, 0); //ssr
455 encode_spectral_coeffs(s, sce);
460 * Write some auxiliary information about the created AAC file.
462 static void put_bitstream_info(AACEncContext *s, const char *name)
464 int i, namelen, padbits;
466 namelen = strlen(name) + 2;
467 put_bits(&s->pb, 3, TYPE_FIL);
468 put_bits(&s->pb, 4, FFMIN(namelen, 15));
470 put_bits(&s->pb, 8, namelen - 14);
471 put_bits(&s->pb, 4, 0); //extension type - filler
472 padbits = -put_bits_count(&s->pb) & 7;
473 avpriv_align_put_bits(&s->pb);
474 for (i = 0; i < namelen - 2; i++)
475 put_bits(&s->pb, 8, name[i]);
476 put_bits(&s->pb, 12 - padbits, 0);
480 * Copy input samples.
481 * Channels are reordered from libavcodec's default order to AAC order.
483 static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
486 int end = 2048 + (frame ? frame->nb_samples : 0);
487 const uint8_t *channel_map = aac_chan_maps[s->channels - 1];
489 /* copy and remap input samples */
490 for (ch = 0; ch < s->channels; ch++) {
491 /* copy last 1024 samples of previous frame to the start of the current frame */
492 memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
494 /* copy new samples and zero any remaining samples */
496 memcpy(&s->planar_samples[ch][2048],
497 frame->extended_data[channel_map[ch]],
498 frame->nb_samples * sizeof(s->planar_samples[0][0]));
500 memset(&s->planar_samples[ch][end], 0,
501 (3072 - end) * sizeof(s->planar_samples[0][0]));
505 static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
506 const AVFrame *frame, int *got_packet_ptr)
508 AACEncContext *s = avctx->priv_data;
509 float **samples = s->planar_samples, *samples2, *la, *overlap;
511 int i, ch, w, g, chans, tag, start_ch, ret;
512 int chan_el_counter[4];
513 FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
515 if (s->last_frame == 2)
518 /* add current frame to queue */
520 if ((ret = ff_af_queue_add(&s->afq, frame) < 0))
524 copy_input_samples(s, frame);
526 ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
528 if (!avctx->frame_number)
532 for (i = 0; i < s->chan_map[0]; i++) {
533 FFPsyWindowInfo* wi = windows + start_ch;
534 tag = s->chan_map[i+1];
535 chans = tag == TYPE_CPE ? 2 : 1;
537 for (ch = 0; ch < chans; ch++) {
538 IndividualChannelStream *ics = &cpe->ch[ch].ics;
539 int cur_channel = start_ch + ch;
540 overlap = &samples[cur_channel][0];
541 samples2 = overlap + 1024;
542 la = samples2 + (448+64);
545 if (tag == TYPE_LFE) {
546 wi[ch].window_type[0] = ONLY_LONG_SEQUENCE;
547 wi[ch].window_shape = 0;
548 wi[ch].num_windows = 1;
549 wi[ch].grouping[0] = 1;
551 /* Only the lowest 12 coefficients are used in a LFE channel.
552 * The expression below results in only the bottom 8 coefficients
553 * being used for 11.025kHz to 16kHz sample rates.
555 ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
557 wi[ch] = s->psy.model->window(&s->psy, samples2, la, cur_channel,
558 ics->window_sequence[0]);
560 ics->window_sequence[1] = ics->window_sequence[0];
561 ics->window_sequence[0] = wi[ch].window_type[0];
562 ics->use_kb_window[1] = ics->use_kb_window[0];
563 ics->use_kb_window[0] = wi[ch].window_shape;
564 ics->num_windows = wi[ch].num_windows;
565 ics->swb_sizes = s->psy.bands [ics->num_windows == 8];
566 ics->num_swb = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
567 for (w = 0; w < ics->num_windows; w++)
568 ics->group_len[w] = wi[ch].grouping[w];
570 apply_window_and_mdct(s, &cpe->ch[ch], overlap);
574 if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels))) {
575 av_log(avctx, AV_LOG_ERROR, "Error getting output packet\n");
581 init_put_bits(&s->pb, avpkt->data, avpkt->size);
583 if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & CODEC_FLAG_BITEXACT))
584 put_bitstream_info(s, LIBAVCODEC_IDENT);
586 memset(chan_el_counter, 0, sizeof(chan_el_counter));
587 for (i = 0; i < s->chan_map[0]; i++) {
588 FFPsyWindowInfo* wi = windows + start_ch;
589 const float *coeffs[2];
590 tag = s->chan_map[i+1];
591 chans = tag == TYPE_CPE ? 2 : 1;
593 put_bits(&s->pb, 3, tag);
594 put_bits(&s->pb, 4, chan_el_counter[tag]++);
595 for (ch = 0; ch < chans; ch++)
596 coeffs[ch] = cpe->ch[ch].coeffs;
597 s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
598 for (ch = 0; ch < chans; ch++) {
599 s->cur_channel = start_ch * 2 + ch;
600 s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
602 cpe->common_window = 0;
604 && wi[0].window_type[0] == wi[1].window_type[0]
605 && wi[0].window_shape == wi[1].window_shape) {
607 cpe->common_window = 1;
608 for (w = 0; w < wi[0].num_windows; w++) {
609 if (wi[0].grouping[w] != wi[1].grouping[w]) {
610 cpe->common_window = 0;
615 s->cur_channel = start_ch * 2;
616 if (s->options.stereo_mode && cpe->common_window) {
617 if (s->options.stereo_mode > 0) {
618 IndividualChannelStream *ics = &cpe->ch[0].ics;
619 for (w = 0; w < ics->num_windows; w += ics->group_len[w])
620 for (g = 0; g < ics->num_swb; g++)
621 cpe->ms_mask[w*16+g] = 1;
622 } else if (s->coder->search_for_ms) {
623 s->coder->search_for_ms(s, cpe, s->lambda);
626 adjust_frame_information(cpe, chans);
628 put_bits(&s->pb, 1, cpe->common_window);
629 if (cpe->common_window) {
630 put_ics_info(s, &cpe->ch[0].ics);
631 encode_ms_info(&s->pb, cpe);
634 for (ch = 0; ch < chans; ch++) {
635 s->cur_channel = start_ch + ch;
636 encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
641 frame_bits = put_bits_count(&s->pb);
642 if (frame_bits <= 6144 * s->channels - 3) {
643 s->psy.bitres.bits = frame_bits / s->channels;
647 s->lambda *= avctx->bit_rate * 1024.0f / avctx->sample_rate / frame_bits;
651 put_bits(&s->pb, 3, TYPE_END);
652 flush_put_bits(&s->pb);
653 avctx->frame_bits = put_bits_count(&s->pb);
655 // rate control stuff
656 if (!(avctx->flags & CODEC_FLAG_QSCALE)) {
657 float ratio = avctx->bit_rate * 1024.0f / avctx->sample_rate / avctx->frame_bits;
659 s->lambda = FFMIN(s->lambda, 65536.f);
665 ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
668 avpkt->size = put_bits_count(&s->pb) >> 3;
673 static av_cold int aac_encode_end(AVCodecContext *avctx)
675 AACEncContext *s = avctx->priv_data;
677 ff_mdct_end(&s->mdct1024);
678 ff_mdct_end(&s->mdct128);
681 ff_psy_preprocess_end(s->psypp);
682 av_freep(&s->buffer.samples);
684 ff_af_queue_close(&s->afq);
685 #if FF_API_OLD_ENCODE_AUDIO
686 av_freep(&avctx->coded_frame);
691 static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
695 ff_dsputil_init(&s->dsp, avctx);
696 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
699 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
700 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
701 ff_init_ff_sine_windows(10);
702 ff_init_ff_sine_windows(7);
704 if (ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0))
706 if (ret = ff_mdct_init(&s->mdct128, 8, 0, 32768.0))
712 static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
715 FF_ALLOCZ_OR_GOTO(avctx, s->buffer.samples, 3 * 1024 * s->channels * sizeof(s->buffer.samples[0]), alloc_fail);
716 FF_ALLOCZ_OR_GOTO(avctx, s->cpe, sizeof(ChannelElement) * s->chan_map[0], alloc_fail);
717 FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + FF_INPUT_BUFFER_PADDING_SIZE, alloc_fail);
719 for(ch = 0; ch < s->channels; ch++)
720 s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
722 #if FF_API_OLD_ENCODE_AUDIO
723 if (!(avctx->coded_frame = avcodec_alloc_frame()))
729 return AVERROR(ENOMEM);
732 static av_cold int aac_encode_init(AVCodecContext *avctx)
734 AACEncContext *s = avctx->priv_data;
736 const uint8_t *sizes[2];
737 uint8_t grouping[AAC_MAX_CHANNELS];
740 avctx->frame_size = 1024;
742 for (i = 0; i < 16; i++)
743 if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
746 s->channels = avctx->channels;
749 "Unsupported sample rate %d\n", avctx->sample_rate);
750 ERROR_IF(s->channels > AAC_MAX_CHANNELS,
751 "Unsupported number of channels: %d\n", s->channels);
752 ERROR_IF(avctx->profile != FF_PROFILE_UNKNOWN && avctx->profile != FF_PROFILE_AAC_LOW,
753 "Unsupported profile %d\n", avctx->profile);
754 ERROR_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
755 "Too many bits per frame requested\n");
757 s->samplerate_index = i;
759 s->chan_map = aac_chan_configs[s->channels-1];
761 if (ret = dsp_init(avctx, s))
764 if (ret = alloc_buffers(avctx, s))
767 avctx->extradata_size = 5;
768 put_audio_specific_config(avctx);
770 sizes[0] = swb_size_1024[i];
771 sizes[1] = swb_size_128[i];
772 lengths[0] = ff_aac_num_swb_1024[i];
773 lengths[1] = ff_aac_num_swb_128[i];
774 for (i = 0; i < s->chan_map[0]; i++)
775 grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
776 if (ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths, s->chan_map[0], grouping))
778 s->psypp = ff_psy_preprocess_init(avctx);
779 s->coder = &ff_aac_coders[s->options.aac_coder];
781 s->lambda = avctx->global_quality ? avctx->global_quality : 120;
785 for (i = 0; i < 428; i++)
786 ff_aac_pow34sf_tab[i] = sqrt(ff_aac_pow2sf_tab[i] * sqrt(ff_aac_pow2sf_tab[i]));
789 ff_af_queue_init(avctx, &s->afq);
793 aac_encode_end(avctx);
797 #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
798 static const AVOption aacenc_options[] = {
799 {"stereo_mode", "Stereo coding method", offsetof(AACEncContext, options.stereo_mode), AV_OPT_TYPE_INT, {.i64 = 0}, -1, 1, AACENC_FLAGS, "stereo_mode"},
800 {"auto", "Selected by the Encoder", 0, AV_OPT_TYPE_CONST, {.i64 = -1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
801 {"ms_off", "Disable Mid/Side coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
802 {"ms_force", "Force Mid/Side for the whole frame if possible", 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
803 {"aac_coder", "", offsetof(AACEncContext, options.aac_coder), AV_OPT_TYPE_INT, {.i64 = 2}, 0, AAC_CODER_NB-1, AACENC_FLAGS},
807 static const AVClass aacenc_class = {
809 av_default_item_name,
811 LIBAVUTIL_VERSION_INT,
814 /* duplicated from avpriv_mpeg4audio_sample_rates to avoid shared build
816 static const int mpeg4audio_sample_rates[16] = {
817 96000, 88200, 64000, 48000, 44100, 32000,
818 24000, 22050, 16000, 12000, 11025, 8000, 7350
821 AVCodec ff_aac_encoder = {
823 .type = AVMEDIA_TYPE_AUDIO,
824 .id = AV_CODEC_ID_AAC,
825 .priv_data_size = sizeof(AACEncContext),
826 .init = aac_encode_init,
827 .encode2 = aac_encode_frame,
828 .close = aac_encode_end,
829 .supported_samplerates = mpeg4audio_sample_rates,
830 .capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY |
831 CODEC_CAP_EXPERIMENTAL,
832 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
833 AV_SAMPLE_FMT_NONE },
834 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
835 .priv_class = &aacenc_class,