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.
77827 Commits
32 Branches
303MB
Tree: 0634c54253
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '0634c54253'
${ noResults }
FFmpeg/libavcodec/aacenc.c

1065 lines
40KB
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/thread.h"

  33. #include "libavutil/float_dsp.h"

  34. #include "libavutil/opt.h"

  35. #include "avcodec.h"

  36. #include "put_bits.h"

  37. #include "internal.h"

  38. #include "mpeg4audio.h"

  39. #include "kbdwin.h"

  40. #include "sinewin.h"

  41. 

  42. #include "aac.h"

  43. #include "aactab.h"

  44. #include "aacenc.h"

  45. #include "aacenctab.h"

  46. #include "aacenc_utils.h"

  47. 

  48. #include "psymodel.h"

  49. 

  50. static AVOnce aac_table_init = AV_ONCE_INIT;

  51. 

  52. /**

  53.  * Make AAC audio config object.

  54.  * @see 1.6.2.1 "Syntax - AudioSpecificConfig"

  55.  */

  56. static void put_audio_specific_config(AVCodecContext *avctx)

  57. {

  58.     PutBitContext pb;

  59.     AACEncContext *s = avctx->priv_data;

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

  61. 

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

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

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

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

  66.     //GASpecificConfig

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

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

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

  70. 

  71.     //Explicitly Mark SBR absent

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

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

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

  75.     flush_put_bits(&pb);

  76. }

  77. 

  78. void ff_quantize_band_cost_cache_init(struct AACEncContext *s)

  79. {

  80.     int sf, g;

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

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

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

  84.         }

  85.     }

  86. }

  87. 

  88. #define WINDOW_FUNC(type) \

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

  90.                                     SingleChannelElement *sce, \

  91.                                     const float *audio)

  92. 

  93. WINDOW_FUNC(only_long)

  94. {

  95.     const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;

  96.     const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;

  97.     float *out = sce->ret_buf;

  98. 

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

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

  101. }

  102. 

  103. WINDOW_FUNC(long_start)

  104. {

  105.     const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;

  106.     const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;

  107.     float *out = sce->ret_buf;

  108. 

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

  110.     memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);

  111.     fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);

  112.     memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);

  113. }

  114. 

  115. WINDOW_FUNC(long_stop)

  116. {

  117.     const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;

  118.     const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;

  119.     float *out = sce->ret_buf;

  120. 

  121.     memset(out, 0, sizeof(out[0]) * 448);

  122.     fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);

  123.     memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);

  124.     fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);

  125. }

  126. 

  127. WINDOW_FUNC(eight_short)

  128. {

  129.     const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;

  130.     const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;

  131.     const float *in = audio + 448;

  132.     float *out = sce->ret_buf;

  133.     int w;

  134. 

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

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

  137.         out += 128;

  138.         in  += 128;

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

  140.         out += 128;

  141.     }

  142. }

  143. 

  144. static void (*const apply_window[4])(AVFloatDSPContext *fdsp,

  145.                                      SingleChannelElement *sce,

  146.                                      const float *audio) = {

  147.     [ONLY_LONG_SEQUENCE]   = apply_only_long_window,

  148.     [LONG_START_SEQUENCE]  = apply_long_start_window,

  149.     [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,

  150.     [LONG_STOP_SEQUENCE]   = apply_long_stop_window

  151. };

  152. 

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

  154.                                   float *audio)

  155. {

  156.     int i;

  157.     float *output = sce->ret_buf;

  158. 

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

  160. 

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

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

  163.     else

  164.         for (i = 0; i < 1024; i += 128)

  165.             s->mdct128.mdct_calc(&s->mdct128, &sce->coeffs[i], output + i*2);

  166.     memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);

  167.     memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));

  168. }

  169. 

  170. /**

  171.  * Encode ics_info element.

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

  173.  */

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

  175. {

  176.     int w;

  177. 

  178.     put_bits(&s->pb, 1, 0);                // ics_reserved bit

  179.     put_bits(&s->pb, 2, info->window_sequence[0]);

  180.     put_bits(&s->pb, 1, info->use_kb_window[0]);

  181.     if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {

  182.         put_bits(&s->pb, 6, info->max_sfb);

  183.         put_bits(&s->pb, 1, !!info->predictor_present);

  184.     } else {

  185.         put_bits(&s->pb, 4, info->max_sfb);

  186.         for (w = 1; w < 8; w++)

  187.             put_bits(&s->pb, 1, !info->group_len[w]);

  188.     }

  189. }

  190. 

  191. /**

  192.  * Encode MS data.

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

  194.  */

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

  196. {

  197.     int i, w;

  198. 

  199.     put_bits(pb, 2, cpe->ms_mode);

  200.     if (cpe->ms_mode == 1)

  201.         for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])

  202.             for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)

  203.                 put_bits(pb, 1, cpe->ms_mask[w*16 + i]);

  204. }

  205. 

  206. /**

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

  208.  */

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

  210. {

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

  212.     int maxsfb, cmaxsfb;

  213. 

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

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

  216.         maxsfb = 0;

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

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

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

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

  221.                     ;

  222.                 maxsfb = FFMAX(maxsfb, cmaxsfb);

  223.             }

  224.         }

  225.         ics->max_sfb = maxsfb;

  226. 

  227.         //adjust zero bands for window groups

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

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

  230.                 i = 1;

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

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

  233.                         i = 0;

  234.                         break;

  235.                     }

  236.                 }

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

  238.             }

  239.         }

  240.     }

  241. 

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

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

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

  245.         int msc = 0;

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

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

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

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

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

  251.                     msc++;

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

  253.             cpe->ms_mode = 0;

  254.         else

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

  256.     }

  257. }

  258. 

  259. static void apply_intensity_stereo(ChannelElement *cpe)

  260. {

  261.     int w, w2, g, i;

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

  263.     if (!cpe->common_window)

  264.         return;

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

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

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

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

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

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

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

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

  273.                     continue;

  274.                 }

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

  276.                     p *= -1;

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

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

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

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

  281.                 }

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

  283.             }

  284.         }

  285.     }

  286. }

  287. 

  288. static void apply_mid_side_stereo(ChannelElement *cpe)

  289. {

  290.     int w, w2, g, i;

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

  292.     if (!cpe->common_window)

  293.         return;

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

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

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

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

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

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

  300.                  * ms_mask is set.

  301.                  */

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

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

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

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

  306.                     continue;

  307.                 }

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

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

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

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

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

  313.                 }

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

  315.             }

  316.         }

  317.     }

  318. }

  319. 

  320. /**

  321.  * Encode scalefactor band coding type.

  322.  */

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

  324. {

  325.     int w;

  326. 

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

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

  329. 

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

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

  332. }

  333. 

  334. /**

  335.  * Encode scalefactors.

  336.  */

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

  338.                                  SingleChannelElement *sce)

  339. {

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

  341.     int off_is = 0, noise_flag = 1;

  342.     int i, w;

  343. 

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

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

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

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

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

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

  350.                     if (noise_flag-- > 0) {

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

  352.                         continue;

  353.                     }

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

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

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

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

  358.                 } else {

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

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

  361.                 }

  362.                 diff += SCALE_DIFF_ZERO;

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

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

  365.             }

  366.         }

  367.     }

  368. }

  369. 

  370. /**

  371.  * Encode pulse data.

  372.  */

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

  374. {

  375.     int i;

  376. 

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

  378.     if (!pulse->num_pulse)

  379.         return;

  380. 

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

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

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

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

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

  386.     }

  387. }

  388. 

  389. /**

  390.  * Encode spectral coefficients processed by psychoacoustic model.

  391.  */

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

  393. {

  394.     int start, i, w, w2;

  395. 

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

  397.         start = 0;

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

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

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

  401.                 continue;

  402.             }

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

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

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

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

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

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

  409.                                                    s->lambda,

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

  411.             }

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

  413.         }

  414.     }

  415. }

  416. 

  417. /**

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

  419.  */

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

  421. {

  422.     int start, i, j, w;

  423. 

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

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

  426.             start = 0;

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

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

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

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

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

  432.             }

  433.         }

  434.     }

  435. }

  436. 

  437. /**

  438.  * Encode one channel of audio data.

  439.  */

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

  441.                                      SingleChannelElement *sce,

  442.                                      int common_window)

  443. {

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

  445.     if (!common_window) {

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

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

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

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

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

  451.     }

  452.     encode_band_info(s, sce);

  453.     encode_scale_factors(avctx, s, sce);

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

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

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

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

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

  459.     encode_spectral_coeffs(s, sce);

  460.     return 0;

  461. }

  462. 

  463. /**

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

  465.  */

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

  467. {

  468.     int i, namelen, padbits;

  469. 

  470.     namelen = strlen(name) + 2;

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

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

  473.     if (namelen >= 15)

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

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

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

  477.     avpriv_align_put_bits(&s->pb);

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

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

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

  481. }

  482. 

  483. /*

  484.  * Copy input samples.

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

  486.  */

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

  488. {

  489.     int ch;

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

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

  492. 

  493.     /* copy and remap input samples */

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

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

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

  497. 

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

  499.         if (frame) {

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

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

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

  503.         }

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

  505.                (3072 - end) * sizeof(s->planar_samples[0][0]));

  506.     }

  507. }

  508. 

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

  510.                             const AVFrame *frame, int *got_packet_ptr)

  511. {

  512.     AACEncContext *s = avctx->priv_data;

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

  514.     ChannelElement *cpe;

  515.     SingleChannelElement *sce;

  516.     IndividualChannelStream *ics;

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

  518.     int target_bits, rate_bits, too_many_bits, too_few_bits;

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

  520.     int chan_el_counter[4];

  521.     FFPsyWindowInfo windows[AAC_MAX_CHANNELS];

  522. 

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

  524.         return 0;

  525. 

  526.     /* add current frame to queue */

  527.     if (frame) {

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

  529.             return ret;

  530.     }

  531. 

  532.     copy_input_samples(s, frame);

  533.     if (s->psypp)

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

  535. 

  536.     if (!avctx->frame_number)

  537.         return 0;

  538. 

  539.     start_ch = 0;

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

  541.         FFPsyWindowInfo* wi = windows + start_ch;

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

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

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

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

  546.             float clip_avoidance_factor;

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

  548.             ics = &sce->ics;

  549.             s->cur_channel = start_ch + ch;

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

  551.             samples2 = overlap + 1024;

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

  553.             if (!frame)

  554.                 la = NULL;

  555.             if (tag == TYPE_LFE) {

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

  557.                 wi[ch].window_shape   = 0;

  558.                 wi[ch].num_windows    = 1;

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

  560. 

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

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

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

  564.                  */

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

  566.             } else {

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

  568.                                               ics->window_sequence[0]);

  569.             }

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

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

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

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

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

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

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

  577.             ics->max_sfb            = FFMIN(ics->max_sfb, ics->num_swb);

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

  579.                                         ff_swb_offset_128 [s->samplerate_index]:

  580.                                         ff_swb_offset_1024[s->samplerate_index];

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

  582.                                         ff_tns_max_bands_128 [s->samplerate_index]:

  583.                                         ff_tns_max_bands_1024[s->samplerate_index];

  584.             clip_avoidance_factor = 0.0f;

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

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

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

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

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

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

  591.                 } else {

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

  593.                 }

  594.             }

  595.             if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {

  596.                 ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;

  597.             } else {

  598.                 ics->clip_avoidance_factor = 1.0f;

  599.             }

  600. 

  601.             apply_window_and_mdct(s, sce, overlap);

  602. 

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

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

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

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

  607.             }

  608. 

  609.             if (isnan(cpe->ch->coeffs[0]) ||

  610.                 isnan(cpe->ch->coeffs[  128]) ||

  611.                 isnan(cpe->ch->coeffs[2*128]) ||

  612.                 isnan(cpe->ch->coeffs[3*128]) ||

  613.                 isnan(cpe->ch->coeffs[4*128]) ||

  614.                 isnan(cpe->ch->coeffs[5*128]) ||

  615.                 isnan(cpe->ch->coeffs[6*128]) ||

  616.                 isnan(cpe->ch->coeffs[7*128])

  617.             ) {

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

  619.                 return AVERROR(EINVAL);

  620.             }

  621.             avoid_clipping(s, sce);

  622.         }

  623.         start_ch += chans;

  624.     }

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

  626.         return ret;

  627.     frame_bits = its = 0;

  628.     do {

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

  630. 

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

  632.             put_bitstream_info(s, LIBAVCODEC_IDENT);

  633.         start_ch = 0;

  634.         target_bits = 0;

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

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

  637.             FFPsyWindowInfo* wi = windows + start_ch;

  638.             const float *coeffs[2];

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

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

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

  642.             cpe->common_window = 0;

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

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

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

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

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

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

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

  650.                 sce->ics.predictor_present = 0;

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

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

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

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

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

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

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

  658.             }

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

  660.             s->psy.bitres.bits = s->last_frame_pb_count / s->channels;

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

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

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

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

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

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

  667.             }

  668.             s->cur_type = tag;

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

  670.                 s->cur_channel = start_ch + ch;

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

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

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

  674.             }

  675.             if (chans > 1

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

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

  678. 

  679.                 cpe->common_window = 1;

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

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

  682.                         cpe->common_window = 0;

  683.                         break;

  684.                     }

  685.                 }

  686.             }

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

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

  689.                 s->cur_channel = start_ch + ch;

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

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

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

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

  694.                 if (sce->tns.present)

  695.                     tns_mode = 1;

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

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

  698.             }

  699.             s->cur_channel = start_ch;

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

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

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

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

  704.                 apply_intensity_stereo(cpe);

  705.             }

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

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

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

  709.                     s->cur_channel = start_ch + ch;

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

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

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

  713.                 }

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

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

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

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

  718.                     s->cur_channel = start_ch + ch;

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

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

  721.                 }

  722.                 s->cur_channel = start_ch;

  723.             }

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

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

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

  727.                 else if (cpe->common_window)

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

  729.                 apply_mid_side_stereo(cpe);

  730.             }

  731.             adjust_frame_information(cpe, chans);

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

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

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

  735.                     s->cur_channel = start_ch + ch;

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

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

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

  739.                 }

  740.                 s->cur_channel = start_ch;

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

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

  743.             }

  744.             if (chans == 2) {

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

  746.                 if (cpe->common_window) {

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

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

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

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

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

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

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

  754.                 }

  755.             }

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

  757.                 s->cur_channel = start_ch + ch;

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

  759.             }

  760.             start_ch += chans;

  761.         }

  762. 

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

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

  765.             break;

  766.         }

  767. 

  768.         /* rate control stuff

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

  770.          * but drift towards the nominal bitrate always

  771.          */

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

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

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

  775.         too_many_bits = FFMAX(target_bits, rate_bits);

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

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

  778. 

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

  780.         too_few_bits = too_few_bits - too_few_bits/8;

  781.         too_many_bits = too_many_bits + too_many_bits/2;

  782. 

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

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

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

  786.         {

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

  788. 

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

  790.                 /*

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

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

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

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

  795.                  */

  796.                 ratio = sqrtf(sqrtf(ratio));

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

  798.             } else {

  799.                 /* Not so fast though */

  800.                 ratio = sqrtf(ratio);

  801.             }

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

  803. 

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

  805.             if (ratio > 0.9f && ratio < 1.1f) {

  806.                 break;

  807.             } else {

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

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

  810.                         // Must restore coeffs

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

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

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

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

  815.                     }

  816.                 }

  817.                 its++;

  818.             }

  819.         } else {

  820.             break;

  821.         }

  822.     } while (1);

  823. 

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

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

  826. 

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

  828.     flush_put_bits(&s->pb);

  829. 

  830.     s->last_frame_pb_count = put_bits_count(&s->pb);

  831. 

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

  833.     s->lambda_count++;

  834. 

  835.     if (!frame)

  836.         s->last_frame++;

  837. 

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

  839.                        &avpkt->duration);

  840. 

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

  842.     *got_packet_ptr = 1;

  843.     return 0;

  844. }

  845. 

  846. static av_cold int aac_encode_end(AVCodecContext *avctx)

  847. {

  848.     AACEncContext *s = avctx->priv_data;

  849. 

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

  851. 

  852.     ff_mdct_end(&s->mdct1024);

  853.     ff_mdct_end(&s->mdct128);

  854.     ff_psy_end(&s->psy);

  855.     ff_lpc_end(&s->lpc);

  856.     if (s->psypp)

  857.         ff_psy_preprocess_end(s->psypp);

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

  859.     av_freep(&s->cpe);

  860.     av_freep(&s->fdsp);

  861.     ff_af_queue_close(&s->afq);

  862.     return 0;

  863. }

  864. 

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

  866. {

  867.     int ret = 0;

  868. 

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

  870.     if (!s->fdsp)

  871.         return AVERROR(ENOMEM);

  872. 

  873.     // window init

  874.     ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);

  875.     ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);

  876.     ff_init_ff_sine_windows(10);

  877.     ff_init_ff_sine_windows(7);

  878. 

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

  880.         return ret;

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

  882.         return ret;

  883. 

  884.     return 0;

  885. }

  886. 

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

  888. {

  889.     int ch;

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

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

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

  893. 

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

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

  896. 

  897.     return 0;

  898. alloc_fail:

  899.     return AVERROR(ENOMEM);

  900. }

  901. 

  902. static av_cold void aac_encode_init_tables(void)

  903. {

  904.     ff_aac_tableinit();

  905. }

  906. 

  907. static av_cold int aac_encode_init(AVCodecContext *avctx)

  908. {

  909.     AACEncContext *s = avctx->priv_data;

  910.     int i, ret = 0;

  911.     const uint8_t *sizes[2];

  912.     uint8_t grouping[AAC_MAX_CHANNELS];

  913.     int lengths[2];

  914. 

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

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

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

  918.     s->last_frame_pb_count = 0;

  919.     avctx->extradata_size = 5;

  920.     avctx->frame_size = 1024;

  921.     avctx->initial_padding = 1024;

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

  923.         6144 * s->channels / 1024.0 * avctx->sample_rate,

  924.         avctx->bit_rate);

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

  926.                      avctx->profile;

  927. 

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

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

  930.             break;

  931.     s->samplerate_index = i;

  932. 

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

  934.              s->samplerate_index >= ff_aac_swb_size_1024_len ||

  935.              s->samplerate_index >= ff_aac_swb_size_128_len,

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

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

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

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

  940.              "Too many bits %f > %d per frame requested, clamping to max\n",

  941.              1024.0 * avctx->bit_rate / avctx->sample_rate,

  942.              6144 * s->channels);

  943. 

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

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

  946.             break;

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

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

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

  950.         avctx->profile = FF_PROFILE_AAC_LOW;

  951.         ERROR_IF(s->options.pred,

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

  953.         ERROR_IF(s->options.ltp,

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

  955.         WARN_IF(s->options.pns,

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

  957.         s->options.pns = 0;

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

  959.         s->options.ltp = 1;

  960.         ERROR_IF(s->options.pred,

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

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

  963.         s->options.pred = 1;

  964.         ERROR_IF(s->options.ltp,

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

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

  967.         avctx->profile = FF_PROFILE_AAC_LTP;

  968.         WARN_IF(1,

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

  970.         ERROR_IF(s->options.pred,

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

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

  973.         avctx->profile = FF_PROFILE_AAC_MAIN;

  974.         WARN_IF(1,

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

  976.         ERROR_IF(s->options.ltp,

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

  978.     }

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

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

  981. 

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

  983.         ERROR_IF(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,

  984.                  "Coders other than twoloop require -strict -2 and some may be removed in the future\n");

  985.         WARN_IF(s->options.coder == AAC_CODER_FAAC,

  986.                 "The FAAC-like coder will be removed in the near future, please use twoloop!\n");

  987.         s->options.intensity_stereo = 0;

  988.         s->options.pns = 0;

  989.     }

  990. 

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

  992.         goto fail;

  993. 

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

  995.         goto fail;

  996. 

  997.     put_audio_specific_config(avctx);

  998. 

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

  1000.     sizes[1]   = ff_aac_swb_size_128[s->samplerate_index];

  1001.     lengths[0] = ff_aac_num_swb_1024[s->samplerate_index];

  1002.     lengths[1] = ff_aac_num_swb_128[s->samplerate_index];

  1003.     for (i = 0; i < s->chan_map[0]; i++)

  1004.         grouping[i] = s->chan_map[i + 1] == TYPE_CPE;

  1005.     if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,

  1006.                            s->chan_map[0], grouping)) < 0)

  1007.         goto fail;

  1008.     s->psypp = ff_psy_preprocess_init(avctx);

  1009.     ff_lpc_init(&s->lpc, 2*avctx->frame_size, TNS_MAX_ORDER, FF_LPC_TYPE_LEVINSON);

  1010.     av_lfg_init(&s->lfg, 0x72adca55);

  1011. 

  1012.     if (HAVE_MIPSDSP)

  1013.         ff_aac_coder_init_mips(s);

  1014. 

  1015.     if ((ret = ff_thread_once(&aac_table_init, &aac_encode_init_tables)) != 0)

  1016.         return AVERROR_UNKNOWN;

  1017. 

  1018.     ff_af_queue_init(avctx, &s->afq);

  1019. 

  1020.     return 0;

  1021. fail:

  1022.     aac_encode_end(avctx);

  1023.     return ret;

  1024. }

  1025. 

  1026. #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM

  1027. static const AVOption aacenc_options[] = {

  1028.     {"aac_coder", "Coding algorithm", offsetof(AACEncContext, options.coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_TWOLOOP}, -1, AAC_CODER_NB-1, AACENC_FLAGS, "coder"},

  1029.         {"faac",     "FAAC-inspired method",      0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAAC},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},

  1030.         {"anmr",     "ANMR method",               0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},

  1031.         {"twoloop",  "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},

  1032.         {"fast",     "Constant quantizer",        0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},

  1033.     {"aac_ms", "Force M/S stereo coding", offsetof(AACEncContext, options.mid_side), AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, AACENC_FLAGS},

  1034.     {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},

  1035.     {"aac_pns", "Perceptual noise substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},

  1036.     {"aac_tns", "Temporal noise shaping", offsetof(AACEncContext, options.tns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},

  1037.     {"aac_ltp", "Long term prediction", offsetof(AACEncContext, options.ltp), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},

  1038.     {"aac_pred", "AAC-Main prediction", offsetof(AACEncContext, options.pred), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},

  1039.     {NULL}

  1040. };

  1041. 

  1042. static const AVClass aacenc_class = {

  1043.     "AAC encoder",

  1044.     av_default_item_name,

  1045.     aacenc_options,

  1046.     LIBAVUTIL_VERSION_INT,

  1047. };

  1048. 

  1049. AVCodec ff_aac_encoder = {

  1050.     .name           = "aac",

  1051.     .long_name      = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),

  1052.     .type           = AVMEDIA_TYPE_AUDIO,

  1053.     .id             = AV_CODEC_ID_AAC,

  1054.     .priv_data_size = sizeof(AACEncContext),

  1055.     .init           = aac_encode_init,

  1056.     .encode2        = aac_encode_frame,

  1057.     .close          = aac_encode_end,

  1058.     .supported_samplerates = mpeg4audio_sample_rates,

  1059.     .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE,

  1060.     .capabilities   = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY,

  1061.     .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,

  1062.                                                      AV_SAMPLE_FMT_NONE },

  1063.     .priv_class     = &aacenc_class,

  1064. };