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.
78412 Commits
32 Branches
293MB
Tree: 0fe0e213c0
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '0fe0e213c0'
${ noResults }
FFmpeg/libavcodec/aacenc.c

1081 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/libm.h"

  33. #include "libavutil/thread.h"

  34. #include "libavutil/float_dsp.h"

  35. #include "libavutil/opt.h"

  36. #include "avcodec.h"

  37. #include "put_bits.h"

  38. #include "internal.h"

  39. #include "mpeg4audio.h"

  40. #include "kbdwin.h"

  41. #include "sinewin.h"

  42. 

  43. #include "aac.h"

  44. #include "aactab.h"

  45. #include "aacenc.h"

  46. #include "aacenctab.h"

  47. #include "aacenc_utils.h"

  48. 

  49. #include "psymodel.h"

  50. 

  51. static AVOnce aac_table_init = AV_ONCE_INIT;

  52. 

  53. /**

  54.  * Make AAC audio config object.

  55.  * @see 1.6.2.1 "Syntax - AudioSpecificConfig"

  56.  */

  57. static void put_audio_specific_config(AVCodecContext *avctx)

  58. {

  59.     PutBitContext pb;

  60.     AACEncContext *s = avctx->priv_data;

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

  62. 

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

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

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

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

  67.     //GASpecificConfig

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

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

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

  71. 

  72.     //Explicitly Mark SBR absent

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

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

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

  76.     flush_put_bits(&pb);

  77. }

  78. 

  79. void ff_quantize_band_cost_cache_init(struct AACEncContext *s)

  80. {

  81.     int sf, g;

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

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

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

  85.         }

  86.     }

  87. }

  88. 

  89. #define WINDOW_FUNC(type) \

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

  91.                                     SingleChannelElement *sce, \

  92.                                     const float *audio)

  93. 

  94. WINDOW_FUNC(only_long)

  95. {

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

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

  98.     float *out = sce->ret_buf;

  99. 

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

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

  102. }

  103. 

  104. WINDOW_FUNC(long_start)

  105. {

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

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

  108.     float *out = sce->ret_buf;

  109. 

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

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

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

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

  114. }

  115. 

  116. WINDOW_FUNC(long_stop)

  117. {

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

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

  120.     float *out = sce->ret_buf;

  121. 

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

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

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

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

  126. }

  127. 

  128. WINDOW_FUNC(eight_short)

  129. {

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

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

  132.     const float *in = audio + 448;

  133.     float *out = sce->ret_buf;

  134.     int w;

  135. 

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

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

  138.         out += 128;

  139.         in  += 128;

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

  141.         out += 128;

  142.     }

  143. }

  144. 

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

  146.                                      SingleChannelElement *sce,

  147.                                      const float *audio) = {

  148.     [ONLY_LONG_SEQUENCE]   = apply_only_long_window,

  149.     [LONG_START_SEQUENCE]  = apply_long_start_window,

  150.     [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,

  151.     [LONG_STOP_SEQUENCE]   = apply_long_stop_window

  152. };

  153. 

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

  155.                                   float *audio)

  156. {

  157.     int i;

  158.     const float *output = sce->ret_buf;

  159. 

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

  161. 

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

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

  164.     else

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

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

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

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

  169. }

  170. 

  171. /**

  172.  * Encode ics_info element.

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

  174.  */

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

  176. {

  177.     int w;

  178. 

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

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

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

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

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

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

  185.     } else {

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

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

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

  189.     }

  190. }

  191. 

  192. /**

  193.  * Encode MS data.

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

  195.  */

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

  197. {

  198.     int i, w;

  199. 

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

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

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

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

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

  205. }

  206. 

  207. /**

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

  209.  */

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

  211. {

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

  213.     int maxsfb, cmaxsfb;

  214. 

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

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

  217.         maxsfb = 0;

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

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

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

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

  222.                     ;

  223.                 maxsfb = FFMAX(maxsfb, cmaxsfb);

  224.             }

  225.         }

  226.         ics->max_sfb = maxsfb;

  227. 

  228.         //adjust zero bands for window groups

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

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

  231.                 i = 1;

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

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

  234.                         i = 0;

  235.                         break;

  236.                     }

  237.                 }

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

  239.             }

  240.         }

  241.     }

  242. 

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

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

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

  246.         int msc = 0;

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

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

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

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

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

  252.                     msc++;

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

  254.             cpe->ms_mode = 0;

  255.         else

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

  257.     }

  258. }

  259. 

  260. static void apply_intensity_stereo(ChannelElement *cpe)

  261. {

  262.     int w, w2, g, i;

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

  264.     if (!cpe->common_window)

  265.         return;

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

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

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

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

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

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

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

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

  274.                     continue;

  275.                 }

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

  277.                     p *= -1;

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

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

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

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

  282.                 }

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

  284.             }

  285.         }

  286.     }

  287. }

  288. 

  289. static void apply_mid_side_stereo(ChannelElement *cpe)

  290. {

  291.     int w, w2, g, i;

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

  293.     if (!cpe->common_window)

  294.         return;

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

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

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

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

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

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

  301.                  * ms_mask is set.

  302.                  */

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

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

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

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

  307.                     continue;

  308.                 }

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

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

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

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

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

  314.                 }

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

  316.             }

  317.         }

  318.     }

  319. }

  320. 

  321. /**

  322.  * Encode scalefactor band coding type.

  323.  */

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

  325. {

  326.     int w;

  327. 

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

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

  330. 

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

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

  333. }

  334. 

  335. /**

  336.  * Encode scalefactors.

  337.  */

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

  339.                                  SingleChannelElement *sce)

  340. {

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

  342.     int off_is = 0, noise_flag = 1;

  343.     int i, w;

  344. 

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

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

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

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

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

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

  351.                     if (noise_flag-- > 0) {

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

  353.                         continue;

  354.                     }

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

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

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

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

  359.                 } else {

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

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

  362.                 }

  363.                 diff += SCALE_DIFF_ZERO;

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

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

  366.             }

  367.         }

  368.     }

  369. }

  370. 

  371. /**

  372.  * Encode pulse data.

  373.  */

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

  375. {

  376.     int i;

  377. 

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

  379.     if (!pulse->num_pulse)

  380.         return;

  381. 

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

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

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

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

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

  387.     }

  388. }

  389. 

  390. /**

  391.  * Encode spectral coefficients processed by psychoacoustic model.

  392.  */

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

  394. {

  395.     int start, i, w, w2;

  396. 

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

  398.         start = 0;

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

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

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

  402.                 continue;

  403.             }

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

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

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

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

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

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

  410.                                                    s->lambda,

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

  412.             }

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

  414.         }

  415.     }

  416. }

  417. 

  418. /**

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

  420.  */

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

  422. {

  423.     int start, i, j, w;

  424. 

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

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

  427.             start = 0;

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

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

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

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

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

  433.             }

  434.         }

  435.     }

  436. }

  437. 

  438. /**

  439.  * Encode one channel of audio data.

  440.  */

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

  442.                                      SingleChannelElement *sce,

  443.                                      int common_window)

  444. {

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

  446.     if (!common_window) {

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

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

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

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

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

  452.     }

  453.     encode_band_info(s, sce);

  454.     encode_scale_factors(avctx, s, sce);

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

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

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

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

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

  460.     encode_spectral_coeffs(s, sce);

  461.     return 0;

  462. }

  463. 

  464. /**

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

  466.  */

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

  468. {

  469.     int i, namelen, padbits;

  470. 

  471.     namelen = strlen(name) + 2;

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

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

  474.     if (namelen >= 15)

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

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

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

  478.     avpriv_align_put_bits(&s->pb);

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

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

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

  482. }

  483. 

  484. /*

  485.  * Copy input samples.

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

  487.  */

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

  489. {

  490.     int ch;

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

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

  493. 

  494.     /* copy and remap input samples */

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

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

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

  498. 

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

  500.         if (frame) {

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

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

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

  504.         }

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

  506.                (3072 - end) * sizeof(s->planar_samples[0][0]));

  507.     }

  508. }

  509. 

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

  511.                             const AVFrame *frame, int *got_packet_ptr)

  512. {

  513.     AACEncContext *s = avctx->priv_data;

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

  515.     ChannelElement *cpe;

  516.     SingleChannelElement *sce;

  517.     IndividualChannelStream *ics;

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

  519.     int target_bits, rate_bits, too_many_bits, too_few_bits;

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

  521.     int chan_el_counter[4];

  522.     FFPsyWindowInfo windows[AAC_MAX_CHANNELS];

  523. 

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

  525.         return 0;

  526. 

  527.     /* add current frame to queue */

  528.     if (frame) {

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

  530.             return ret;

  531.     }

  532. 

  533.     copy_input_samples(s, frame);

  534.     if (s->psypp)

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

  536. 

  537.     if (!avctx->frame_number)

  538.         return 0;

  539. 

  540.     start_ch = 0;

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

  542.         FFPsyWindowInfo* wi = windows + start_ch;

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

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

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

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

  547.             int k;

  548.             float clip_avoidance_factor;

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

  550.             ics = &sce->ics;

  551.             s->cur_channel = start_ch + ch;

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

  553.             samples2 = overlap + 1024;

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

  555.             if (!frame)

  556.                 la = NULL;

  557.             if (tag == TYPE_LFE) {

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

  559.                 wi[ch].window_shape   = 0;

  560.                 wi[ch].num_windows    = 1;

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

  562. 

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

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

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

  566.                  */

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

  568.             } else {

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

  570.                                               ics->window_sequence[0]);

  571.             }

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

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

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

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

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

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

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

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

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

  581.                                         ff_swb_offset_128 [s->samplerate_index]:

  582.                                         ff_swb_offset_1024[s->samplerate_index];

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

  584.                                         ff_tns_max_bands_128 [s->samplerate_index]:

  585.                                         ff_tns_max_bands_1024[s->samplerate_index];

  586.             clip_avoidance_factor = 0.0f;

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

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

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

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

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

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

  593.                 } else {

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

  595.                 }

  596.             }

  597.             if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {

  598.                 ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;

  599.             } else {

  600.                 ics->clip_avoidance_factor = 1.0f;

  601.             }

  602. 

  603.             apply_window_and_mdct(s, sce, overlap);

  604. 

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

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

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

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

  609.             }

  610. 

  611.             for (k = 0; k < 1024; k++) {

  612.                 if (!isfinite(cpe->ch[ch].coeffs[k])) {

  613.                     av_log(avctx, AV_LOG_ERROR, "Input contains NaN/+-Inf\n");

  614.                     return AVERROR(EINVAL);

  615.                 }

  616.             }

  617.             avoid_clipping(s, sce);

  618.         }

  619.         start_ch += chans;

  620.     }

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

  622.         return ret;

  623.     frame_bits = its = 0;

  624.     do {

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

  626. 

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

  628.             put_bitstream_info(s, LIBAVCODEC_IDENT);

  629.         start_ch = 0;

  630.         target_bits = 0;

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

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

  633.             FFPsyWindowInfo* wi = windows + start_ch;

  634.             const float *coeffs[2];

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

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

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

  638.             cpe->common_window = 0;

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

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

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

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

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

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

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

  646.                 sce->ics.predictor_present = 0;

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

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

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

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

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

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

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

  654.             }

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

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

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

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

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

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

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

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

  663.             }

  664.             s->cur_type = tag;

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

  666.                 s->cur_channel = start_ch + ch;

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

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

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

  670.             }

  671.             if (chans > 1

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

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

  674. 

  675.                 cpe->common_window = 1;

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

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

  678.                         cpe->common_window = 0;

  679.                         break;

  680.                     }

  681.                 }

  682.             }

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

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

  685.                 s->cur_channel = start_ch + ch;

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

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

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

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

  690.                 if (sce->tns.present)

  691.                     tns_mode = 1;

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

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

  694.             }

  695.             s->cur_channel = start_ch;

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

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

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

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

  700.                 apply_intensity_stereo(cpe);

  701.             }

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

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

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

  705.                     s->cur_channel = start_ch + ch;

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

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

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

  709.                 }

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

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

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

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

  714.                     s->cur_channel = start_ch + ch;

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

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

  717.                 }

  718.                 s->cur_channel = start_ch;

  719.             }

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

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

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

  723.                 else if (cpe->common_window)

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

  725.                 apply_mid_side_stereo(cpe);

  726.             }

  727.             adjust_frame_information(cpe, chans);

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

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

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

  731.                     s->cur_channel = start_ch + ch;

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

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

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

  735.                 }

  736.                 s->cur_channel = start_ch;

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

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

  739.             }

  740.             if (chans == 2) {

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

  742.                 if (cpe->common_window) {

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

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

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

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

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

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

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

  750.                 }

  751.             }

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

  753.                 s->cur_channel = start_ch + ch;

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

  755.             }

  756.             start_ch += chans;

  757.         }

  758. 

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

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

  761.             break;

  762.         }

  763. 

  764.         /* rate control stuff

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

  766.          * but drift towards the nominal bitrate always

  767.          */

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

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

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

  771.         too_many_bits = FFMAX(target_bits, rate_bits);

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

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

  774. 

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

  776.         too_few_bits = too_few_bits - too_few_bits/8;

  777.         too_many_bits = too_many_bits + too_many_bits/2;

  778. 

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

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

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

  782.         {

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

  784. 

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

  786.                 /*

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

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

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

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

  791.                  */

  792.                 ratio = sqrtf(sqrtf(ratio));

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

  794.             } else {

  795.                 /* Not so fast though */

  796.                 ratio = sqrtf(ratio);

  797.             }

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

  799. 

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

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

  802.                 break;

  803.             } else {

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

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

  806.                         // Must restore coeffs

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

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

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

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

  811.                     }

  812.                 }

  813.                 its++;

  814.             }

  815.         } else {

  816.             break;

  817.         }

  818.     } while (1);

  819. 

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

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

  822. 

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

  824.     flush_put_bits(&s->pb);

  825. 

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

  827. 

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

  829.     s->lambda_count++;

  830. 

  831.     if (!frame)

  832.         s->last_frame++;

  833. 

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

  835.                        &avpkt->duration);

  836. 

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

  838.     *got_packet_ptr = 1;

  839.     return 0;

  840. }

  841. 

  842. static av_cold int aac_encode_end(AVCodecContext *avctx)

  843. {

  844.     AACEncContext *s = avctx->priv_data;

  845. 

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

  847. 

  848.     ff_mdct_end(&s->mdct1024);

  849.     ff_mdct_end(&s->mdct128);

  850.     ff_psy_end(&s->psy);

  851.     ff_lpc_end(&s->lpc);

  852.     if (s->psypp)

  853.         ff_psy_preprocess_end(s->psypp);

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

  855.     av_freep(&s->cpe);

  856.     av_freep(&s->fdsp);

  857.     ff_af_queue_close(&s->afq);

  858.     return 0;

  859. }

  860. 

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

  862. {

  863.     int ret = 0;

  864. 

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

  866.     if (!s->fdsp)

  867.         return AVERROR(ENOMEM);

  868. 

  869.     // window init

  870.     ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);

  871.     ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);

  872.     ff_init_ff_sine_windows(10);

  873.     ff_init_ff_sine_windows(7);

  874. 

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

  876.         return ret;

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

  878.         return ret;

  879. 

  880.     return 0;

  881. }

  882. 

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

  884. {

  885.     int ch;

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

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

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

  889. 

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

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

  892. 

  893.     return 0;

  894. alloc_fail:

  895.     return AVERROR(ENOMEM);

  896. }

  897. 

  898. static av_cold void aac_encode_init_tables(void)

  899. {

  900.     ff_aac_tableinit();

  901. }

  902. 

  903. static av_cold int aac_encode_init(AVCodecContext *avctx)

  904. {

  905.     AACEncContext *s = avctx->priv_data;

  906.     int i, ret = 0;

  907.     const uint8_t *sizes[2];

  908.     uint8_t grouping[AAC_MAX_CHANNELS];

  909.     int lengths[2];

  910. 

  911.     /* Constants */

  912.     s->last_frame_pb_count = 0;

  913.     avctx->extradata_size = 5;

  914.     avctx->frame_size = 1024;

  915.     avctx->initial_padding = 1024;

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

  917. 

  918.     /* Channel map and unspecified bitrate guessing */

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

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

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

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

  923.     if (!avctx->bit_rate) {

  924.         for (i = 1; i <= s->chan_map[0]; i++) {

  925.             avctx->bit_rate += s->chan_map[i] == TYPE_CPE ? 128000 : /* Pair */

  926.                                s->chan_map[i] == TYPE_LFE ? 16000  : /* LFE  */

  927.                                                             69000  ; /* SCE  */

  928.         }

  929.     }

  930. 

  931.     /* Samplerate */

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

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

  934.             break;

  935.     s->samplerate_index = i;

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

  937.              s->samplerate_index >= ff_aac_swb_size_1024_len ||

  938.              s->samplerate_index >= ff_aac_swb_size_128_len,

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

  940. 

  941.     /* Bitrate limiting */

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

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

  944.              1024.0 * avctx->bit_rate / avctx->sample_rate,

  945.              6144 * s->channels);

  946.     avctx->bit_rate = (int64_t)FFMIN(6144 * s->channels / 1024.0 * avctx->sample_rate,

  947.                                      avctx->bit_rate);

  948. 

  949.     /* Profile and option setting */

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

  951.                      avctx->profile;

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

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

  954.             break;

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

  956.         avctx->profile = FF_PROFILE_AAC_LOW;

  957.         ERROR_IF(s->options.pred,

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

  959.         ERROR_IF(s->options.ltp,

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

  961.         WARN_IF(s->options.pns,

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

  963.         s->options.pns = 0;

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

  965.         s->options.ltp = 1;

  966.         ERROR_IF(s->options.pred,

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

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

  969.         s->options.pred = 1;

  970.         ERROR_IF(s->options.ltp,

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

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

  973.         avctx->profile = FF_PROFILE_AAC_LTP;

  974.         WARN_IF(1,

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

  976.         ERROR_IF(s->options.pred,

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

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

  979.         avctx->profile = FF_PROFILE_AAC_MAIN;

  980.         WARN_IF(1,

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

  982.         ERROR_IF(s->options.ltp,

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

  984.     }

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

  986. 

  987.     /* Coder limitations */

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

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

  990.         ERROR_IF(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,

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

  992.         s->options.intensity_stereo = 0;

  993.         s->options.pns = 0;

  994.     }

  995.     ERROR_IF(s->options.ltp && avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL,

  996.              "The LPT profile requires experimental compliance, add -strict -2 to enable!\n");

  997. 

  998.     /* M/S introduces horrible artifacts with multichannel files, this is temporary */

  999.     if (s->channels > 3)

  1000.         s->options.mid_side = 0;

  1001. 

  1002.     if ((ret = dsp_init(avctx, s)) < 0)

  1003.         goto fail;

  1004. 

  1005.     if ((ret = alloc_buffers(avctx, s)) < 0)

  1006.         goto fail;

  1007. 

  1008.     put_audio_specific_config(avctx);

  1009. 

  1010.     sizes[0]   = ff_aac_swb_size_1024[s->samplerate_index];

  1011.     sizes[1]   = ff_aac_swb_size_128[s->samplerate_index];

  1012.     lengths[0] = ff_aac_num_swb_1024[s->samplerate_index];

  1013.     lengths[1] = ff_aac_num_swb_128[s->samplerate_index];

  1014.     for (i = 0; i < s->chan_map[0]; i++)

  1015.         grouping[i] = s->chan_map[i + 1] == TYPE_CPE;

  1016.     if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,

  1017.                            s->chan_map[0], grouping)) < 0)

  1018.         goto fail;

  1019.     s->psypp = ff_psy_preprocess_init(avctx);

  1020.     ff_lpc_init(&s->lpc, 2*avctx->frame_size, TNS_MAX_ORDER, FF_LPC_TYPE_LEVINSON);

  1021.     av_lfg_init(&s->lfg, 0x72adca55);

  1022. 

  1023.     if (HAVE_MIPSDSP)

  1024.         ff_aac_coder_init_mips(s);

  1025. 

  1026.     if ((ret = ff_thread_once(&aac_table_init, &aac_encode_init_tables)) != 0)

  1027.         return AVERROR_UNKNOWN;

  1028. 

  1029.     ff_af_queue_init(avctx, &s->afq);

  1030. 

  1031.     return 0;

  1032. fail:

  1033.     aac_encode_end(avctx);

  1034.     return ret;

  1035. }

  1036. 

  1037. #define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM

  1038. static const AVOption aacenc_options[] = {

  1039.     {"aac_coder", "Coding algorithm", offsetof(AACEncContext, options.coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_TWOLOOP}, 0, AAC_CODER_NB-1, AACENC_FLAGS, "coder"},

  1040.         {"anmr",     "ANMR method",               0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},

  1041.         {"twoloop",  "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},

  1042.         {"fast",     "Constant quantizer",        0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST},    INT_MIN, INT_MAX, AACENC_FLAGS, "coder"},

  1043.     {"aac_ms", "Force M/S stereo coding", offsetof(AACEncContext, options.mid_side), AV_OPT_TYPE_BOOL, {.i64 = -1}, -1, 1, AACENC_FLAGS},

  1044.     {"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},

  1045.     {"aac_pns", "Perceptual noise substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},

  1046.     {"aac_tns", "Temporal noise shaping", offsetof(AACEncContext, options.tns), AV_OPT_TYPE_BOOL, {.i64 = 1}, -1, 1, AACENC_FLAGS},

  1047.     {"aac_ltp", "Long term prediction", offsetof(AACEncContext, options.ltp), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},

  1048.     {"aac_pred", "AAC-Main prediction", offsetof(AACEncContext, options.pred), AV_OPT_TYPE_BOOL, {.i64 = 0}, -1, 1, AACENC_FLAGS},

  1049.     {NULL}

  1050. };

  1051. 

  1052. static const AVClass aacenc_class = {

  1053.     "AAC encoder",

  1054.     av_default_item_name,

  1055.     aacenc_options,

  1056.     LIBAVUTIL_VERSION_INT,

  1057. };

  1058. 

  1059. static const AVCodecDefault aac_encode_defaults[] = {

  1060.     { "b", "0" },

  1061.     { NULL }

  1062. };

  1063. 

  1064. AVCodec ff_aac_encoder = {

  1065.     .name           = "aac",

  1066.     .long_name      = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),

  1067.     .type           = AVMEDIA_TYPE_AUDIO,

  1068.     .id             = AV_CODEC_ID_AAC,

  1069.     .priv_data_size = sizeof(AACEncContext),

  1070.     .init           = aac_encode_init,

  1071.     .encode2        = aac_encode_frame,

  1072.     .close          = aac_encode_end,

  1073.     .defaults       = aac_encode_defaults,

  1074.     .supported_samplerates = mpeg4audio_sample_rates,

  1075.     .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE,

  1076.     .capabilities   = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY,

  1077.     .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,

  1078.                                                      AV_SAMPLE_FMT_NONE },

  1079.     .priv_class     = &aacenc_class,

  1080. };