falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
76849 Commits
32 Branches
462MB
Tree: fc36d852ee
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'fc36d852ee'
${ noResults }
FFmpeg/libavcodec/aacenc.c

1038 lines
39KB
Raw Blame History 

  1. /*

  2.  * AAC encoder

  3.  * Copyright (C) 2008 Konstantin Shishkov

  4.  *

  5.  * This file is part of FFmpeg.

  6.  *

  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.

  11.  *

  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.

  16.  *

  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

  20.  */

  21. 

  22. /**

  23.  * @file

  24.  * AAC encoder

  25.  */

  26. 

  27. /***********************************

  28.  *              TODOs:

  29.  * add sane pulse detection

  30.  ***********************************/

  31. 

  32. #include "libavutil/float_dsp.h"

  33. #include "libavutil/opt.h"

  34. #include "avcodec.h"

  35. #include "put_bits.h"

  36. #include "internal.h"

  37. #include "mpeg4audio.h"

  38. #include "kbdwin.h"

  39. #include "sinewin.h"

  40. 

  41. #include "aac.h"

  42. #include "aactab.h"

  43. #include "aacenc.h"

  44. #include "aacenctab.h"

  45. #include "aacenc_utils.h"

  46. 

  47. #include "psymodel.h"

  48. 

  49. /**

  50.  * Make AAC audio config object.

  51.  * @see 1.6.2.1 "Syntax - AudioSpecificConfig"

  52.  */

  53. static void put_audio_specific_config(AVCodecContext *avctx)

  54. {

  55.     PutBitContext pb;

  56.     AACEncContext *s = avctx->priv_data;

  57.     int channels = s->channels - (s->channels == 8 ? 1 : 0);

  58. 

  59.     init_put_bits(&pb, avctx->extradata, avctx->extradata_size);

  60.     put_bits(&pb, 5, s->profile+1); //profile

  61.     put_bits(&pb, 4, s->samplerate_index); //sample rate index

  62.     put_bits(&pb, 4, channels);

  63.     //GASpecificConfig

  64.     put_bits(&pb, 1, 0); //frame length - 1024 samples

  65.     put_bits(&pb, 1, 0); //does not depend on core coder

  66.     put_bits(&pb, 1, 0); //is not extension

  67. 

  68.     //Explicitly Mark SBR absent

  69.     put_bits(&pb, 11, 0x2b7); //sync extension

  70.     put_bits(&pb, 5,  AOT_SBR);

  71.     put_bits(&pb, 1,  0);

  72.     flush_put_bits(&pb);

  73. }

  74. 

  75. void ff_quantize_band_cost_cache_init(struct AACEncContext *s)

  76. {

  77.     int sf, g;

  78.     for (sf = 0; sf < 256; sf++) {

  79.         for (g = 0; g < 128; g++) {

  80.             s->quantize_band_cost_cache[sf][g].bits = -1;

  81.         }

  82.     }

  83. }

  84. 

  85. #define WINDOW_FUNC(type) \

  86. static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \

  87.                                     SingleChannelElement *sce, \

  88.                                     const float *audio)

  89. 

  90. WINDOW_FUNC(only_long)

  91. {

  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;

  95. 

  96.     fdsp->vector_fmul        (out,        audio,        lwindow, 1024);

  97.     fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);

  98. }

  99. 

  100. WINDOW_FUNC(long_start)

  101. {

  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;

  105. 

  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);

  110. }

  111. 

  112. WINDOW_FUNC(long_stop)

  113. {

  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;

  117. 

  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);

  122. }

  123. 

  124. WINDOW_FUNC(eight_short)

  125. {

  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;

  130.     int w;

  131. 

  132.     for (w = 0; w < 8; w++) {

  133.         fdsp->vector_fmul        (out, in, w ? pwindow : swindow, 128);

  134.         out += 128;

  135.         in  += 128;

  136.         fdsp->vector_fmul_reverse(out, in, swindow, 128);

  137.         out += 128;

  138.     }

  139. }

  140. 

  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

  148. };

  149. 

  150. static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,

  151.                                   float *audio)

  152. {

  153.     int i;

  154.     float *output = sce->ret_buf;

  155. 

  156.     apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);

  157. 

  158.     if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)

  159.         s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);

  160.     else

  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));

  165. }

  166. 

  167. /**

  168.  * Encode ics_info element.

  169.  * @see Table 4.6 (syntax of ics_info)

  170.  */

  171. static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)

  172. {

  173.     int w;

  174. 

  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, !!info->predictor_present);

  181.     } else {

  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]);

  185.     }

  186. }

  187. 

  188. /**

  189.  * Encode MS data.

  190.  * @see 4.6.8.1 "Joint Coding - M/S Stereo"

  191.  */

  192. static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)

  193. {

  194.     int i, w;

  195. 

  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]);

  201. }

  202. 

  203. /**

  204.  * Produce integer coefficients from scalefactors provided by the model.

  205.  */

  206. static void adjust_frame_information(ChannelElement *cpe, int chans)

  207. {

  208.     int i, w, w2, g, ch;

  209.     int maxsfb, cmaxsfb;

  210. 

  211.     for (ch = 0; ch < chans; ch++) {

  212.         IndividualChannelStream *ics = &cpe->ch[ch].ics;

  213.         maxsfb = 0;

  214.         cpe->ch[ch].pulse.num_pulse = 0;

  215.         for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {

  216.             for (w2 =  0; w2 < ics->group_len[w]; w2++) {

  217.                 for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)

  218.                     ;

  219.                 maxsfb = FFMAX(maxsfb, cmaxsfb);

  220.             }

  221.         }

  222.         ics->max_sfb = maxsfb;

  223. 

  224.         //adjust zero bands for window groups

  225.         for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {

  226.             for (g = 0; g < ics->max_sfb; g++) {

  227.                 i = 1;

  228.                 for (w2 = w; w2 < w + ics->group_len[w]; w2++) {

  229.                     if (!cpe->ch[ch].zeroes[w2*16 + g]) {

  230.                         i = 0;

  231.                         break;

  232.                     }

  233.                 }

  234.                 cpe->ch[ch].zeroes[w*16 + g] = i;

  235.             }

  236.         }

  237.     }

  238. 

  239.     if (chans > 1 && cpe->common_window) {

  240.         IndividualChannelStream *ics0 = &cpe->ch[0].ics;

  241.         IndividualChannelStream *ics1 = &cpe->ch[1].ics;

  242.         int msc = 0;

  243.         ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);

  244.         ics1->max_sfb = ics0->max_sfb;

  245.         for (w = 0; w < ics0->num_windows*16; w += 16)

  246.             for (i = 0; i < ics0->max_sfb; i++)

  247.                 if (cpe->ms_mask[w+i])

  248.                     msc++;

  249.         if (msc == 0 || ics0->max_sfb == 0)

  250.             cpe->ms_mode = 0;

  251.         else

  252.             cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;

  253.     }

  254. }

  255. 

  256. static void apply_intensity_stereo(ChannelElement *cpe)

  257. {

  258.     int w, w2, g, i;

  259.     IndividualChannelStream *ics = &cpe->ch[0].ics;

  260.     if (!cpe->common_window)

  261.         return;

  262.     for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {

  263.         for (w2 =  0; w2 < ics->group_len[w]; w2++) {

  264.             int start = (w+w2) * 128;

  265.             for (g = 0; g < ics->num_swb; g++) {

  266.                 int p  = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);

  267.                 float scale = cpe->ch[0].is_ener[w*16+g];

  268.                 if (!cpe->is_mask[w*16 + g]) {

  269.                     start += ics->swb_sizes[g];

  270.                     continue;

  271.                 }

  272.                 if (cpe->ms_mask[w*16 + g])

  273.                     p *= -1;

  274.                 for (i = 0; i < ics->swb_sizes[g]; i++) {

  275.                     float sum = (cpe->ch[0].coeffs[start+i] + p*cpe->ch[1].coeffs[start+i])*scale;

  276.                     cpe->ch[0].coeffs[start+i] = sum;

  277.                     cpe->ch[1].coeffs[start+i] = 0.0f;

  278.                 }

  279.                 start += ics->swb_sizes[g];

  280.             }

  281.         }

  282.     }

  283. }

  284. 

  285. static void apply_mid_side_stereo(ChannelElement *cpe)

  286. {

  287.     int w, w2, g, i;

  288.     IndividualChannelStream *ics = &cpe->ch[0].ics;

  289.     if (!cpe->common_window)

  290.         return;

  291.     for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {

  292.         for (w2 =  0; w2 < ics->group_len[w]; w2++) {

  293.             int start = (w+w2) * 128;

  294.             for (g = 0; g < ics->num_swb; g++) {

  295.                 /* ms_mask can be used for other purposes in PNS and I/S,

  296.                  * so must not apply M/S if any band uses either, even if

  297.                  * ms_mask is set.

  298.                  */

  299.                 if (!cpe->ms_mask[w*16 + g] || cpe->is_mask[w*16 + g]

  300.                     || cpe->ch[0].band_type[w*16 + g] == NOISE_BT

  301.                     || cpe->ch[1].band_type[w*16 + g] == NOISE_BT) {

  302.                     start += ics->swb_sizes[g];

  303.                     continue;

  304.                 }

  305.                 for (i = 0; i < ics->swb_sizes[g]; i++) {

  306.                     float L = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) * 0.5f;

  307.                     float R = L - cpe->ch[1].coeffs[start+i];

  308.                     cpe->ch[0].coeffs[start+i] = L;

  309.                     cpe->ch[1].coeffs[start+i] = R;

  310.                 }

  311.                 start += ics->swb_sizes[g];

  312.             }

  313.         }

  314.     }

  315. }

  316. 

  317. /**

  318.  * Encode scalefactor band coding type.

  319.  */

  320. static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)

  321. {

  322.     int w;

  323. 

  324.     if (s->coder->set_special_band_scalefactors)

  325.         s->coder->set_special_band_scalefactors(s, sce);

  326. 

  327.     for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])

  328.         s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);

  329. }

  330. 

  331. /**

  332.  * Encode scalefactors.

  333.  */

  334. static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,

  335.                                  SingleChannelElement *sce)

  336. {

  337.     int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;

  338.     int off_is = 0, noise_flag = 1;

  339.     int i, w;

  340. 

  341.     for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {

  342.         for (i = 0; i < sce->ics.max_sfb; i++) {

  343.             if (!sce->zeroes[w*16 + i]) {

  344.                 if (sce->band_type[w*16 + i] == NOISE_BT) {

  345.                     diff = sce->sf_idx[w*16 + i] - off_pns;

  346.                     off_pns = sce->sf_idx[w*16 + i];

  347.                     if (noise_flag-- > 0) {

  348.                         put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);

  349.                         continue;

  350.                     }

  351.                 } else if (sce->band_type[w*16 + i] == INTENSITY_BT  ||

  352.                            sce->band_type[w*16 + i] == INTENSITY_BT2) {

  353.                     diff = sce->sf_idx[w*16 + i] - off_is;

  354.                     off_is = sce->sf_idx[w*16 + i];

  355.                 } else {

  356.                     diff = sce->sf_idx[w*16 + i] - off_sf;

  357.                     off_sf = sce->sf_idx[w*16 + i];

  358.                 }

  359.                 diff += SCALE_DIFF_ZERO;

  360.                 av_assert0(diff >= 0 && diff <= 120);

  361.                 put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);

  362.             }

  363.         }

  364.     }

  365. }

  366. 

  367. /**

  368.  * Encode pulse data.

  369.  */

  370. static void encode_pulses(AACEncContext *s, Pulse *pulse)

  371. {

  372.     int i;

  373. 

  374.     put_bits(&s->pb, 1, !!pulse->num_pulse);

  375.     if (!pulse->num_pulse)

  376.         return;

  377. 

  378.     put_bits(&s->pb, 2, pulse->num_pulse - 1);

  379.     put_bits(&s->pb, 6, pulse->start);

  380.     for (i = 0; i < pulse->num_pulse; i++) {

  381.         put_bits(&s->pb, 5, pulse->pos[i]);

  382.         put_bits(&s->pb, 4, pulse->amp[i]);

  383.     }

  384. }

  385. 

  386. /**

  387.  * Encode spectral coefficients processed by psychoacoustic model.

  388.  */

  389. static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)

  390. {

  391.     int start, i, w, w2;

  392. 

  393.     for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {

  394.         start = 0;

  395.         for (i = 0; i < sce->ics.max_sfb; i++) {

  396.             if (sce->zeroes[w*16 + i]) {

  397.                 start += sce->ics.swb_sizes[i];

  398.                 continue;

  399.             }

  400.             for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) {

  401.                 s->coder->quantize_and_encode_band(s, &s->pb,

  402.                                                    &sce->coeffs[start + w2*128],

  403.                                                    NULL, sce->ics.swb_sizes[i],

  404.                                                    sce->sf_idx[w*16 + i],

  405.                                                    sce->band_type[w*16 + i],

  406.                                                    s->lambda,

  407.                                                    sce->ics.window_clipping[w]);

  408.             }

  409.             start += sce->ics.swb_sizes[i];

  410.         }

  411.     }

  412. }

  413. 

  414. /**

  415.  * Downscale spectral coefficients for near-clipping windows to avoid artifacts

  416.  */

  417. static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)

  418. {

  419.     int start, i, j, w;

  420. 

  421.     if (sce->ics.clip_avoidance_factor < 1.0f) {

  422.         for (w = 0; w < sce->ics.num_windows; w++) {

  423.             start = 0;

  424.             for (i = 0; i < sce->ics.max_sfb; i++) {

  425.                 float *swb_coeffs = &sce->coeffs[start + w*128];

  426.                 for (j = 0; j < sce->ics.swb_sizes[i]; j++)

  427.                     swb_coeffs[j] *= sce->ics.clip_avoidance_factor;

  428.                 start += sce->ics.swb_sizes[i];

  429.             }

  430.         }

  431.     }

  432. }

  433. 

  434. /**

  435.  * Encode one channel of audio data.

  436.  */

  437. static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,

  438.                                      SingleChannelElement *sce,

  439.                                      int common_window)

  440. {

  441.     put_bits(&s->pb, 8, sce->sf_idx[0]);

  442.     if (!common_window) {

  443.         put_ics_info(s, &sce->ics);

  444.         if (s->coder->encode_main_pred)

  445.             s->coder->encode_main_pred(s, sce);

  446.         if (s->coder->encode_ltp_info)

  447.             s->coder->encode_ltp_info(s, sce, 0);

  448.     }

  449.     encode_band_info(s, sce);

  450.     encode_scale_factors(avctx, s, sce);

  451.     encode_pulses(s, &sce->pulse);

  452.     put_bits(&s->pb, 1, !!sce->tns.present);

  453.     if (s->coder->encode_tns_info)

  454.         s->coder->encode_tns_info(s, sce);

  455.     put_bits(&s->pb, 1, 0); //ssr

  456.     encode_spectral_coeffs(s, sce);

  457.     return 0;

  458. }

  459. 

  460. /**

  461.  * Write some auxiliary information about the created AAC file.

  462.  */

  463. static void put_bitstream_info(AACEncContext *s, const char *name)

  464. {

  465.     int i, namelen, padbits;

  466. 

  467.     namelen = strlen(name) + 2;

  468.     put_bits(&s->pb, 3, TYPE_FIL);

  469.     put_bits(&s->pb, 4, FFMIN(namelen, 15));

  470.     if (namelen >= 15)

  471.         put_bits(&s->pb, 8, namelen - 14);

  472.     put_bits(&s->pb, 4, 0); //extension type - filler

  473.     padbits = -put_bits_count(&s->pb) & 7;

  474.     avpriv_align_put_bits(&s->pb);

  475.     for (i = 0; i < namelen - 2; i++)

  476.         put_bits(&s->pb, 8, name[i]);

  477.     put_bits(&s->pb, 12 - padbits, 0);

  478. }

  479. 

  480. /*

  481.  * Copy input samples.

  482.  * Channels are reordered from libavcodec's default order to AAC order.

  483.  */

  484. static void copy_input_samples(AACEncContext *s, const AVFrame *frame)

  485. {

  486.     int ch;

  487.     int end = 2048 + (frame ? frame->nb_samples : 0);

  488.     const uint8_t *channel_map = aac_chan_maps[s->channels - 1];

  489. 

  490.     /* copy and remap input samples */

  491.     for (ch = 0; ch < s->channels; ch++) {

  492.         /* copy last 1024 samples of previous frame to the start of the current frame */

  493.         memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));

  494. 

  495.         /* copy new samples and zero any remaining samples */

  496.         if (frame) {

  497.             memcpy(&s->planar_samples[ch][2048],

  498.                    frame->extended_data[channel_map[ch]],

  499.                    frame->nb_samples * sizeof(s->planar_samples[0][0]));

  500.         }

  501.         memset(&s->planar_samples[ch][end], 0,

  502.                (3072 - end) * sizeof(s->planar_samples[0][0]));

  503.     }

  504. }

  505. 

  506. static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,

  507.                             const AVFrame *frame, int *got_packet_ptr)

  508. {

  509.     AACEncContext *s = avctx->priv_data;

  510.     float **samples = s->planar_samples, *samples2, *la, *overlap;

  511.     ChannelElement *cpe;

  512.     SingleChannelElement *sce;

  513.     IndividualChannelStream *ics;

  514.     int i, its, ch, w, chans, tag, start_ch, ret, frame_bits;

  515.     int target_bits, rate_bits, too_many_bits, too_few_bits;

  516.     int ms_mode = 0, is_mode = 0, tns_mode = 0, pred_mode = 0;

  517.     int chan_el_counter[4];

  518.     FFPsyWindowInfo windows[AAC_MAX_CHANNELS];

  519. 

  520.     if (s->last_frame == 2)

  521.         return 0;

  522. 

  523.     /* add current frame to queue */

  524.     if (frame) {

  525.         if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)

  526.             return ret;

  527.     }

  528. 

  529.     copy_input_samples(s, frame);

  530.     if (s->psypp)

  531.         ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);

  532. 

  533.     if (!avctx->frame_number)

  534.         return 0;

  535. 

  536.     start_ch = 0;

  537.     for (i = 0; i < s->chan_map[0]; i++) {

  538.         FFPsyWindowInfo* wi = windows + start_ch;

  539.         tag      = s->chan_map[i+1];

  540.         chans    = tag == TYPE_CPE ? 2 : 1;

  541.         cpe      = &s->cpe[i];

  542.         for (ch = 0; ch < chans; ch++) {

  543.             float clip_avoidance_factor;

  544.             sce = &cpe->ch[ch];

  545.             ics = &sce->ics;

  546.             s->cur_channel = start_ch + ch;

  547.             overlap  = &samples[s->cur_channel][0];

  548.             samples2 = overlap + 1024;

  549.             la       = samples2 + (448+64);

  550.             if (!frame)

  551.                 la = NULL;

  552.             if (tag == TYPE_LFE) {

  553.                 wi[ch].window_type[0] = ONLY_LONG_SEQUENCE;

  554.                 wi[ch].window_shape   = 0;

  555.                 wi[ch].num_windows    = 1;

  556.                 wi[ch].grouping[0]    = 1;

  557. 

  558.                 /* Only the lowest 12 coefficients are used in a LFE channel.

  559.                  * The expression below results in only the bottom 8 coefficients

  560.                  * being used for 11.025kHz to 16kHz sample rates.

  561.                  */

  562.                 ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;

  563.             } else {

  564.                 wi[ch] = s->psy.model->window(&s->psy, samples2, la, s->cur_channel,

  565.                                               ics->window_sequence[0]);

  566.             }

  567.             ics->window_sequence[1] = ics->window_sequence[0];

  568.             ics->window_sequence[0] = wi[ch].window_type[0];

  569.             ics->use_kb_window[1]   = ics->use_kb_window[0];

  570.             ics->use_kb_window[0]   = wi[ch].window_shape;

  571.             ics->num_windows        = wi[ch].num_windows;

  572.             ics->swb_sizes          = s->psy.bands    [ics->num_windows == 8];

  573.             ics->num_swb            = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];

  574.             ics->swb_offset         = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?

  575.                                         ff_swb_offset_128 [s->samplerate_index]:

  576.                                         ff_swb_offset_1024[s->samplerate_index];

  577.             ics->tns_max_bands      = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?

  578.                                         ff_tns_max_bands_128 [s->samplerate_index]:

  579.                                         ff_tns_max_bands_1024[s->samplerate_index];

  580.             clip_avoidance_factor = 0.0f;

  581.             for (w = 0; w < ics->num_windows; w++)

  582.                 ics->group_len[w] = wi[ch].grouping[w];

  583.             for (w = 0; w < ics->num_windows; w++) {

  584.                 if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {

  585.                     ics->window_clipping[w] = 1;

  586.                     clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);

  587.                 } else {

  588.                     ics->window_clipping[w] = 0;

  589.                 }

  590.             }

  591.             if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {

  592.                 ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;

  593.             } else {

  594.                 ics->clip_avoidance_factor = 1.0f;

  595.             }

  596. 

  597.             apply_window_and_mdct(s, sce, overlap);

  598. 

  599.             if (s->options.ltp && s->coder->update_ltp) {

  600.                 s->coder->update_ltp(s, sce);

  601.                 apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, &sce->ltp_state[0]);

  602.                 s->mdct1024.mdct_calc(&s->mdct1024, sce->lcoeffs, sce->ret_buf);

  603.             }

  604. 

  605.             if (isnan(cpe->ch->coeffs[0])) {

  606.                 av_log(avctx, AV_LOG_ERROR, "Input contains NaN\n");

  607.                 return AVERROR(EINVAL);

  608.             }

  609.             avoid_clipping(s, sce);

  610.         }

  611.         start_ch += chans;

  612.     }

  613.     if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)

  614.         return ret;

  615.     frame_bits = its = 0;

  616.     do {

  617.         init_put_bits(&s->pb, avpkt->data, avpkt->size);

  618. 

  619.         if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))

  620.             put_bitstream_info(s, LIBAVCODEC_IDENT);

  621.         start_ch = 0;

  622.         target_bits = 0;

  623.         memset(chan_el_counter, 0, sizeof(chan_el_counter));

  624.         for (i = 0; i < s->chan_map[0]; i++) {

  625.             FFPsyWindowInfo* wi = windows + start_ch;

  626.             const float *coeffs[2];

  627.             tag      = s->chan_map[i+1];

  628.             chans    = tag == TYPE_CPE ? 2 : 1;

  629.             cpe      = &s->cpe[i];

  630.             cpe->common_window = 0;

  631.             memset(cpe->is_mask, 0, sizeof(cpe->is_mask));

  632.             memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));

  633.             put_bits(&s->pb, 3, tag);

  634.             put_bits(&s->pb, 4, chan_el_counter[tag]++);

  635.             for (ch = 0; ch < chans; ch++) {

  636.                 sce = &cpe->ch[ch];

  637.                 coeffs[ch] = sce->coeffs;

  638.                 sce->ics.predictor_present = 0;

  639.                 sce->ics.ltp.present = 0;

  640.                 memset(sce->ics.ltp.used, 0, sizeof(sce->ics.ltp.used));

  641.                 memset(sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));

  642.                 memset(&sce->tns, 0, sizeof(TemporalNoiseShaping));

  643.                 for (w = 0; w < 128; w++)

  644.                     if (sce->band_type[w] > RESERVED_BT)

  645.                         sce->band_type[w] = 0;

  646.             }

  647.             s->psy.bitres.alloc = -1;

  648.             s->psy.bitres.bits = avctx->frame_bits / s->channels;

  649.             s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);

  650.             if (s->psy.bitres.alloc > 0) {

  651.                 /* Lambda unused here on purpose, we need to take psy's unscaled allocation */

  652.                 target_bits += s->psy.bitres.alloc

  653.                     * (s->lambda / (avctx->global_quality ? avctx->global_quality : 120));

  654.                 s->psy.bitres.alloc /= chans;

  655.             }

  656.             s->cur_type = tag;

  657.             for (ch = 0; ch < chans; ch++) {

  658.                 s->cur_channel = start_ch + ch;

  659.                 if (s->options.pns && s->coder->mark_pns)

  660.                     s->coder->mark_pns(s, avctx, &cpe->ch[ch]);

  661.                 s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);

  662.             }

  663.             if (chans > 1

  664.                 && wi[0].window_type[0] == wi[1].window_type[0]

  665.                 && wi[0].window_shape   == wi[1].window_shape) {

  666. 

  667.                 cpe->common_window = 1;

  668.                 for (w = 0; w < wi[0].num_windows; w++) {

  669.                     if (wi[0].grouping[w] != wi[1].grouping[w]) {

  670.                         cpe->common_window = 0;

  671.                         break;

  672.                     }

  673.                 }

  674.             }

  675.             for (ch = 0; ch < chans; ch++) { /* TNS and PNS */

  676.                 sce = &cpe->ch[ch];

  677.                 s->cur_channel = start_ch + ch;

  678.                 if (s->options.pns && s->coder->search_for_pns)

  679.                     s->coder->search_for_pns(s, avctx, sce);

  680.                 if (s->options.tns && s->coder->search_for_tns)

  681.                     s->coder->search_for_tns(s, sce);

  682.                 if (s->options.tns && s->coder->apply_tns_filt)

  683.                     s->coder->apply_tns_filt(s, sce);

  684.                 if (sce->tns.present)

  685.                     tns_mode = 1;

  686.             }

  687.             s->cur_channel = start_ch;

  688.             if (s->options.intensity_stereo) { /* Intensity Stereo */

  689.                 if (s->coder->search_for_is)

  690.                     s->coder->search_for_is(s, avctx, cpe);

  691.                 if (cpe->is_mode) is_mode = 1;

  692.                 apply_intensity_stereo(cpe);

  693.             }

  694.             if (s->options.pred) { /* Prediction */

  695.                 for (ch = 0; ch < chans; ch++) {

  696.                     sce = &cpe->ch[ch];

  697.                     s->cur_channel = start_ch + ch;

  698.                     if (s->options.pred && s->coder->search_for_pred)

  699.                         s->coder->search_for_pred(s, sce);

  700.                     if (cpe->ch[ch].ics.predictor_present) pred_mode = 1;

  701.                 }

  702.                 if (s->coder->adjust_common_pred)

  703.                     s->coder->adjust_common_pred(s, cpe);

  704.                 for (ch = 0; ch < chans; ch++) {

  705.                     sce = &cpe->ch[ch];

  706.                     s->cur_channel = start_ch + ch;

  707.                     if (s->options.pred && s->coder->apply_main_pred)

  708.                         s->coder->apply_main_pred(s, sce);

  709.                 }

  710.                 s->cur_channel = start_ch;

  711.             }

  712.             if (s->options.mid_side) { /* Mid/Side stereo */

  713.                 if (s->options.mid_side == -1 && s->coder->search_for_ms)

  714.                     s->coder->search_for_ms(s, cpe);

  715.                 else if (cpe->common_window)

  716.                     memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask));

  717.                 apply_mid_side_stereo(cpe);

  718.             }

  719.             adjust_frame_information(cpe, chans);

  720.             if (s->options.ltp) { /* LTP */

  721.                 for (ch = 0; ch < chans; ch++) {

  722.                     sce = &cpe->ch[ch];

  723.                     s->cur_channel = start_ch + ch;

  724.                     if (s->coder->search_for_ltp)

  725.                         s->coder->search_for_ltp(s, sce, cpe->common_window);

  726.                     if (sce->ics.ltp.present) pred_mode = 1;

  727.                 }

  728.                 s->cur_channel = start_ch;

  729.                 if (s->coder->adjust_common_ltp)

  730.                     s->coder->adjust_common_ltp(s, cpe);

  731.             }

  732.             if (chans == 2) {

  733.                 put_bits(&s->pb, 1, cpe->common_window);

  734.                 if (cpe->common_window) {

  735.                     put_ics_info(s, &cpe->ch[0].ics);

  736.                     if (s->coder->encode_main_pred)

  737.                         s->coder->encode_main_pred(s, &cpe->ch[0]);

  738.                     if (s->coder->encode_ltp_info)

  739.                         s->coder->encode_ltp_info(s, &cpe->ch[0], 1);

  740.                     encode_ms_info(&s->pb, cpe);

  741.                     if (cpe->ms_mode) ms_mode = 1;

  742.                 }

  743.             }

  744.             for (ch = 0; ch < chans; ch++) {

  745.                 s->cur_channel = start_ch + ch;

  746.                 encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);

  747.             }

  748.             start_ch += chans;

  749.         }

  750. 

  751.         if (avctx->flags & CODEC_FLAG_QSCALE) {

  752.             /* When using a constant Q-scale, don't mess with lambda */

  753.             break;

  754.         }

  755. 

  756.         /* rate control stuff

  757.          * allow between the nominal bitrate, and what psy's bit reservoir says to target

  758.          * but drift towards the nominal bitrate always

  759.          */

  760.         frame_bits = put_bits_count(&s->pb);

  761.         rate_bits = avctx->bit_rate * 1024 / avctx->sample_rate;

  762.         rate_bits = FFMIN(rate_bits, 6144 * s->channels - 3);

  763.         too_many_bits = FFMAX(target_bits, rate_bits);

  764.         too_many_bits = FFMIN(too_many_bits, 6144 * s->channels - 3);

  765.         too_few_bits = FFMIN(FFMAX(rate_bits - rate_bits/4, target_bits), too_many_bits);

  766. 

  767.         /* When using ABR, be strict (but only for increasing) */

  768.         too_few_bits = too_few_bits - too_few_bits/8;

  769.         too_many_bits = too_many_bits + too_many_bits/2;

  770. 

  771.         if (   its == 0 /* for steady-state Q-scale tracking */

  772.             || (its < 5 && (frame_bits < too_few_bits || frame_bits > too_many_bits))

  773.             || frame_bits >= 6144 * s->channels - 3  )

  774.         {

  775.             float ratio = ((float)rate_bits) / frame_bits;

  776. 

  777.             if (frame_bits >= too_few_bits && frame_bits <= too_many_bits) {

  778.                 /*

  779.                  * This path is for steady-state Q-scale tracking

  780.                  * When frame bits fall within the stable range, we still need to adjust

  781.                  * lambda to maintain it like so in a stable fashion (large jumps in lambda

  782.                  * create artifacts and should be avoided), but slowly

  783.                  */

  784.                 ratio = sqrtf(sqrtf(ratio));

  785.                 ratio = av_clipf(ratio, 0.9f, 1.1f);

  786.             } else {

  787.                 /* Not so fast though */

  788.                 ratio = sqrtf(ratio);

  789.             }

  790.             s->lambda = FFMIN(s->lambda * ratio, 65536.f);

  791. 

  792.             /* Keep iterating if we must reduce and lambda is in the sky */

  793.             if ((s->lambda < 300.f || ratio > 0.9f) && (s->lambda > 10.f || ratio < 1.1f)) {

  794.                 break;

  795.             } else {

  796.                 if (is_mode || ms_mode || tns_mode || pred_mode) {

  797.                     for (i = 0; i < s->chan_map[0]; i++) {

  798.                         // Must restore coeffs

  799.                         chans = tag == TYPE_CPE ? 2 : 1;

  800.                         cpe = &s->cpe[i];

  801.                         for (ch = 0; ch < chans; ch++)

  802.                             memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));

  803.                     }

  804.                 }

  805.                 its++;

  806.             }

  807.         } else {

  808.             break;

  809.         }

  810.     } while (1);

  811. 

  812.     if (s->options.ltp && s->coder->ltp_insert_new_frame)

  813.         s->coder->ltp_insert_new_frame(s);

  814. 

  815.     put_bits(&s->pb, 3, TYPE_END);

  816.     flush_put_bits(&s->pb);

  817.     avctx->frame_bits = put_bits_count(&s->pb);

  818.     s->lambda_sum += s->lambda;

  819.     s->lambda_count++;

  820. 

  821.     if (!frame)

  822.         s->last_frame++;

  823. 

  824.     ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,

  825.                        &avpkt->duration);

  826. 

  827.     avpkt->size = put_bits_count(&s->pb) >> 3;

  828.     *got_packet_ptr = 1;

  829.     return 0;

  830. }

  831. 

  832. static av_cold int aac_encode_end(AVCodecContext *avctx)

  833. {

  834.     AACEncContext *s = avctx->priv_data;

  835. 

  836.     av_log(avctx, AV_LOG_INFO, "Qavg: %.3f\n", s->lambda_sum / s->lambda_count);

  837. 

  838.     ff_mdct_end(&s->mdct1024);

  839.     ff_mdct_end(&s->mdct128);

  840.     ff_psy_end(&s->psy);

  841.     ff_lpc_end(&s->lpc);

  842.     if (s->psypp)

  843.         ff_psy_preprocess_end(s->psypp);

  844.     av_freep(&s->buffer.samples);

  845.     av_freep(&s->cpe);

  846.     av_freep(&s->fdsp);

  847.     ff_af_queue_close(&s->afq);

  848.     return 0;

  849. }

  850. 

  851. static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)

  852. {

  853.     int ret = 0;

  854. 

  855.     s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);

  856.     if (!s->fdsp)

  857.         return AVERROR(ENOMEM);

  858. 

  859.     // window init

  860.     ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);

  861.     ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);

  862.     ff_init_ff_sine_windows(10);

  863.     ff_init_ff_sine_windows(7);

  864. 

  865.     if ((ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) < 0)

  866.         return ret;

  867.     if ((ret = ff_mdct_init(&s->mdct128,   8, 0, 32768.0)) < 0)

  868.         return ret;

  869. 

  870.     return 0;

  871. }

  872. 

  873. static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)

  874. {

  875.     int ch;

  876.     FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->buffer.samples, s->channels, 3 * 1024 * sizeof(s->buffer.samples[0]), alloc_fail);

  877.     FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->cpe, s->chan_map[0], sizeof(ChannelElement), alloc_fail);

  878.     FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + AV_INPUT_BUFFER_PADDING_SIZE, alloc_fail);

  879. 

  880.     for(ch = 0; ch < s->channels; ch++)

  881.         s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;

  882. 

  883.     return 0;

  884. alloc_fail:

  885.     return AVERROR(ENOMEM);

  886. }

  887. 

  888. static av_cold int aac_encode_init(AVCodecContext *avctx)

  889. {

  890.     AACEncContext *s = avctx->priv_data;

  891.     int i, ret = 0;

  892.     const uint8_t *sizes[2];

  893.     uint8_t grouping[AAC_MAX_CHANNELS];

  894.     int lengths[2];

  895. 

  896.     s->channels = avctx->channels;

  897.     s->chan_map = aac_chan_configs[s->channels-1];

  898.     s->random_state = 0x1f2e3d4c;

  899.     s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;

  900.     avctx->extradata_size = 5;

  901.     avctx->frame_size = 1024;

  902.     avctx->initial_padding = 1024;

  903.     avctx->bit_rate = (int)FFMIN(

  904.         6144 * s->channels / 1024.0 * avctx->sample_rate,

  905.         avctx->bit_rate);

  906.     avctx->profile = avctx->profile == FF_PROFILE_UNKNOWN ? FF_PROFILE_AAC_LOW :

  907.                      avctx->profile;

  908. 

  909.     for (i = 0; i < 16; i++)

  910.         if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])

  911.             break;

  912.     s->samplerate_index = i;

  913. 

  914.     ERROR_IF(s->samplerate_index == 16 ||

  915.              s->samplerate_index >= ff_aac_swb_size_1024_len ||

  916.              s->samplerate_index >= ff_aac_swb_size_128_len,

  917.              "Unsupported sample rate %d\n", avctx->sample_rate);

  918.     ERROR_IF(s->channels > AAC_MAX_CHANNELS || s->channels == 7,

  919.              "Unsupported number of channels: %d\n", s->channels);

  920.     WARN_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,

  921.              "Too many bits per frame requested, clamping to max\n");

  922. 

  923.     for (i = 0; i < FF_ARRAY_ELEMS(aacenc_profiles); i++)

  924.         if (avctx->profile == aacenc_profiles[i])

  925.             break;

  926.     ERROR_IF(i == FF_ARRAY_ELEMS(aacenc_profiles),

  927.              "Unsupported encoding profile: %d\n", avctx->profile);

  928.     if (avctx->profile == FF_PROFILE_MPEG2_AAC_LOW) {

  929.         avctx->profile = FF_PROFILE_AAC_LOW;

  930.         ERROR_IF(s->options.pred,

  931.                  "Main prediction unavailable in the \"mpeg2_aac_low\" profile\n");

  932.         ERROR_IF(s->options.ltp,

  933.                  "LTP prediction unavailable in the \"mpeg2_aac_low\" profile\n");

  934.         WARN_IF(s->options.pns,

  935.                 "PNS unavailable in the \"mpeg2_aac_low\" profile, turning off\n");

  936.         s->options.pns = 0;

  937.     } else if (avctx->profile == FF_PROFILE_AAC_LTP) {

  938.         s->options.ltp = 1;

  939.         ERROR_IF(s->options.pred,

  940.                  "Main prediction unavailable in the \"aac_ltp\" profile\n");

  941.     } else if (avctx->profile == FF_PROFILE_AAC_MAIN) {

  942.         s->options.pred = 1;

  943.         ERROR_IF(s->options.ltp,

  944.                  "LTP prediction unavailable in the \"aac_main\" profile\n");

  945.     } else if (s->options.ltp) {

  946.         avctx->profile = FF_PROFILE_AAC_LTP;

  947.         WARN_IF(1,

  948.                 "Chainging profile to \"aac_ltp\"\n");

  949.         ERROR_IF(s->options.pred,

  950.                  "Main prediction unavailable in the \"aac_ltp\" profile\n");

  951.     } else if (s->options.pred) {

  952.         avctx->profile = FF_PROFILE_AAC_MAIN;

  953.         WARN_IF(1,

  954.                 "Chainging profile to \"aac_main\"\n");

  955.         ERROR_IF(s->options.pred,

  956.                  "LTP prediction unavailable in the \"aac_main\" profile\n");

  957.     }

  958.     s->profile = avctx->profile;

  959.     s->coder = &ff_aac_coders[s->options.coder];

  960. 

  961.     if (s->options.coder != AAC_CODER_TWOLOOP) {

  962.         s->options.intensity_stereo = 0;

  963.         s->options.pns = 0;

  964.     }

  965. 

  966.     if ((ret = dsp_init(avctx, s)) < 0)

  967.         goto fail;

  968. 

  969.     if ((ret = alloc_buffers(avctx, s)) < 0)

  970.         goto fail;

  971. 

  972.     put_audio_specific_config(avctx);

  973. 

  974.     sizes[0]   = ff_aac_swb_size_1024[s->samplerate_index];

  975.     sizes[1]   = ff_aac_swb_size_128[s->samplerate_index];

  976.     lengths[0] = ff_aac_num_swb_1024[s->samplerate_index];

  977.     lengths[1] = ff_aac_num_swb_128[s->samplerate_index];

  978.     for (i = 0; i < s->chan_map[0]; i++)

  979.         grouping[i] = s->chan_map[i + 1] == TYPE_CPE;

  980.     if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,

  981.                            s->chan_map[0], grouping)) < 0)

  982.         goto fail;

  983.     s->psypp = ff_psy_preprocess_init(avctx);

  984.     ff_lpc_init(&s->lpc, 2*avctx->frame_size, TNS_MAX_ORDER, FF_LPC_TYPE_LEVINSON);

  985. 

  986.     if (HAVE_MIPSDSPR1)

  987.         ff_aac_coder_init_mips(s);

  988. 

  989.     ff_aac_tableinit();

  990. 

  991.     ff_af_queue_init(avctx, &s->afq);

  992. 

  993.     return 0;

  994. fail:

  995.     aac_encode_end(avctx);

  996.     return ret;

  997. }

  998. 

  999. #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM

  1000. static const AVOption aacenc_options[] = {

  1001.     {"aac_coder", "Coding algorithm", offsetof(AACEncContext, options.coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_TWOLOOP}, -1, AAC_CODER_NB-1, AACENC_FLAGS, "coder"},

  1002.         {"faac",     "FAAC-inspired method",      0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAAC},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},

  1003.         {"anmr",     "ANMR method",               0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},

  1004.         {"twoloop",  "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},

  1005.         {"fast",     "Constant quantizer",        0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},

  1006.     {"aac_ms", "Force M/S stereo coding", offsetof(AACEncContext, options.mid_side), AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, AACENC_FLAGS},

  1007.     {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},

  1008.     {"aac_pns", "Perceptual noise substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},

  1009.     {"aac_tns", "Temporal noise shaping", offsetof(AACEncContext, options.tns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},

  1010.     {"aac_ltp", "Long term prediction", offsetof(AACEncContext, options.ltp), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},

  1011.     {"aac_pred", "AAC-Main prediction", offsetof(AACEncContext, options.pred), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},

  1012.     {NULL}

  1013. };

  1014. 

  1015. static const AVClass aacenc_class = {

  1016.     "AAC encoder",

  1017.     av_default_item_name,

  1018.     aacenc_options,

  1019.     LIBAVUTIL_VERSION_INT,

  1020. };

  1021. 

  1022. AVCodec ff_aac_encoder = {

  1023.     .name           = "aac",

  1024.     .long_name      = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),

  1025.     .type           = AVMEDIA_TYPE_AUDIO,

  1026.     .id             = AV_CODEC_ID_AAC,

  1027.     .priv_data_size = sizeof(AACEncContext),

  1028.     .init           = aac_encode_init,

  1029.     .encode2        = aac_encode_frame,

  1030.     .close          = aac_encode_end,

  1031.     .supported_samplerates = mpeg4audio_sample_rates,

  1032.     .capabilities   = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY |

  1033.                       AV_CODEC_CAP_EXPERIMENTAL,

  1034.     .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,

  1035.                                                      AV_SAMPLE_FMT_NONE },

  1036.     .priv_class     = &aacenc_class,

  1037. };