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.
100520 Commits
32 Branches
1012MB
Tree: 3693e2fcee
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '3693e2fcee'
${ noResults }
FFmpeg/libavcodec/aacps.c

748 lines
29KB
Raw Blame History 

  1. /*

  2.  * MPEG-4 Parametric Stereo decoding functions

  3.  * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com>

  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.  * Note: Rounding-to-nearest used unless otherwise stated

  22.  *

  23.  */

  24. 

  25. #include <stdint.h>

  26. #include "libavutil/common.h"

  27. #include "libavutil/mathematics.h"

  28. #include "avcodec.h"

  29. #include "aacps.h"

  30. #if USE_FIXED

  31. #include "aacps_fixed_tablegen.h"

  32. #else

  33. #include "libavutil/internal.h"

  34. #include "aacps_tablegen.h"

  35. #endif /* USE_FIXED */

  36. 

  37. static const INTFLOAT g1_Q2[] = {

  38.     Q31(0.0f),  Q31(0.01899487526049f), Q31(0.0f), Q31(-0.07293139167538f),

  39.     Q31(0.0f),  Q31(0.30596630545168f), Q31(0.5f)

  40. };

  41. 

  42. static void ipdopd_reset(int8_t *ipd_hist, int8_t *opd_hist)

  43. {

  44.     int i;

  45.     for (i = 0; i < PS_MAX_NR_IPDOPD; i++) {

  46.         opd_hist[i] = 0;

  47.         ipd_hist[i] = 0;

  48.     }

  49. }

  50. 

  51. /** Split one subband into 2 subsubbands with a symmetric real filter.

  52.  * The filter must have its non-center even coefficients equal to zero. */

  53. static void hybrid2_re(INTFLOAT (*in)[2], INTFLOAT (*out)[32][2], const INTFLOAT filter[8], int len, int reverse)

  54. {

  55.     int i, j;

  56.     for (i = 0; i < len; i++, in++) {

  57.         INT64FLOAT re_in = AAC_MUL31(filter[6], in[6][0]); //real inphase

  58.         INT64FLOAT re_op = 0.0f;                          //real out of phase

  59.         INT64FLOAT im_in = AAC_MUL31(filter[6], in[6][1]); //imag inphase

  60.         INT64FLOAT im_op = 0.0f;                          //imag out of phase

  61.         for (j = 0; j < 6; j += 2) {

  62.             re_op += (INT64FLOAT)filter[j+1] * (in[j+1][0] + in[12-j-1][0]);

  63.             im_op += (INT64FLOAT)filter[j+1] * (in[j+1][1] + in[12-j-1][1]);

  64.         }

  65. 

  66. #if USE_FIXED

  67.         re_op = (re_op + 0x40000000) >> 31;

  68.         im_op = (im_op + 0x40000000) >> 31;

  69. #endif /* USE_FIXED */

  70. 

  71.         out[ reverse][i][0] = (INTFLOAT)(re_in + re_op);

  72.         out[ reverse][i][1] = (INTFLOAT)(im_in + im_op);

  73.         out[!reverse][i][0] = (INTFLOAT)(re_in - re_op);

  74.         out[!reverse][i][1] = (INTFLOAT)(im_in - im_op);

  75.     }

  76. }

  77. 

  78. /** Split one subband into 6 subsubbands with a complex filter */

  79. static void hybrid6_cx(PSDSPContext *dsp, INTFLOAT (*in)[2], INTFLOAT (*out)[32][2],

  80.                        TABLE_CONST INTFLOAT (*filter)[8][2], int len)

  81. {

  82.     int i;

  83.     int N = 8;

  84.     LOCAL_ALIGNED_16(INTFLOAT, temp, [8], [2]);

  85. 

  86.     for (i = 0; i < len; i++, in++) {

  87.         dsp->hybrid_analysis(temp, in, (const INTFLOAT (*)[8][2]) filter, 1, N);

  88.         out[0][i][0] = temp[6][0];

  89.         out[0][i][1] = temp[6][1];

  90.         out[1][i][0] = temp[7][0];

  91.         out[1][i][1] = temp[7][1];

  92.         out[2][i][0] = temp[0][0];

  93.         out[2][i][1] = temp[0][1];

  94.         out[3][i][0] = temp[1][0];

  95.         out[3][i][1] = temp[1][1];

  96.         out[4][i][0] = temp[2][0] + temp[5][0];

  97.         out[4][i][1] = temp[2][1] + temp[5][1];

  98.         out[5][i][0] = temp[3][0] + temp[4][0];

  99.         out[5][i][1] = temp[3][1] + temp[4][1];

  100.     }

  101. }

  102. 

  103. static void hybrid4_8_12_cx(PSDSPContext *dsp,

  104.                             INTFLOAT (*in)[2], INTFLOAT (*out)[32][2],

  105.                             TABLE_CONST INTFLOAT (*filter)[8][2], int N, int len)

  106. {

  107.     int i;

  108. 

  109.     for (i = 0; i < len; i++, in++) {

  110.         dsp->hybrid_analysis(out[0] + i, in, (const INTFLOAT (*)[8][2]) filter, 32, N);

  111.     }

  112. }

  113. 

  114. static void hybrid_analysis(PSDSPContext *dsp, INTFLOAT out[91][32][2],

  115.                             INTFLOAT in[5][44][2], INTFLOAT L[2][38][64],

  116.                             int is34, int len)

  117. {

  118.     int i, j;

  119.     for (i = 0; i < 5; i++) {

  120.         for (j = 0; j < 38; j++) {

  121.             in[i][j+6][0] = L[0][j][i];

  122.             in[i][j+6][1] = L[1][j][i];

  123.         }

  124.     }

  125.     if (is34) {

  126.         hybrid4_8_12_cx(dsp, in[0], out,    f34_0_12, 12, len);

  127.         hybrid4_8_12_cx(dsp, in[1], out+12, f34_1_8,   8, len);

  128.         hybrid4_8_12_cx(dsp, in[2], out+20, f34_2_4,   4, len);

  129.         hybrid4_8_12_cx(dsp, in[3], out+24, f34_2_4,   4, len);

  130.         hybrid4_8_12_cx(dsp, in[4], out+28, f34_2_4,   4, len);

  131.         dsp->hybrid_analysis_ileave(out + 27, L, 5, len);

  132.     } else {

  133.         hybrid6_cx(dsp, in[0], out, f20_0_8, len);

  134.         hybrid2_re(in[1], out+6, g1_Q2, len, 1);

  135.         hybrid2_re(in[2], out+8, g1_Q2, len, 0);

  136.         dsp->hybrid_analysis_ileave(out + 7, L, 3, len);

  137.     }

  138.     //update in_buf

  139.     for (i = 0; i < 5; i++) {

  140.         memcpy(in[i], in[i]+32, 6 * sizeof(in[i][0]));

  141.     }

  142. }

  143. 

  144. static void hybrid_synthesis(PSDSPContext *dsp, INTFLOAT out[2][38][64],

  145.                              INTFLOAT in[91][32][2], int is34, int len)

  146. {

  147.     int i, n;

  148.     if (is34) {

  149.         for (n = 0; n < len; n++) {

  150.             memset(out[0][n], 0, 5*sizeof(out[0][n][0]));

  151.             memset(out[1][n], 0, 5*sizeof(out[1][n][0]));

  152.             for (i = 0; i < 12; i++) {

  153.                 out[0][n][0] += (UINTFLOAT)in[   i][n][0];

  154.                 out[1][n][0] += (UINTFLOAT)in[   i][n][1];

  155.             }

  156.             for (i = 0; i < 8; i++) {

  157.                 out[0][n][1] += (UINTFLOAT)in[12+i][n][0];

  158.                 out[1][n][1] += (UINTFLOAT)in[12+i][n][1];

  159.             }

  160.             for (i = 0; i < 4; i++) {

  161.                 out[0][n][2] += (UINTFLOAT)in[20+i][n][0];

  162.                 out[1][n][2] += (UINTFLOAT)in[20+i][n][1];

  163.                 out[0][n][3] += (UINTFLOAT)in[24+i][n][0];

  164.                 out[1][n][3] += (UINTFLOAT)in[24+i][n][1];

  165.                 out[0][n][4] += (UINTFLOAT)in[28+i][n][0];

  166.                 out[1][n][4] += (UINTFLOAT)in[28+i][n][1];

  167.             }

  168.         }

  169.         dsp->hybrid_synthesis_deint(out, in + 27, 5, len);

  170.     } else {

  171.         for (n = 0; n < len; n++) {

  172.             out[0][n][0] = (UINTFLOAT)in[0][n][0] + in[1][n][0] + in[2][n][0] +

  173.                            (UINTFLOAT)in[3][n][0] + in[4][n][0] + in[5][n][0];

  174.             out[1][n][0] = (UINTFLOAT)in[0][n][1] + in[1][n][1] + in[2][n][1] +

  175.                            (UINTFLOAT)in[3][n][1] + in[4][n][1] + in[5][n][1];

  176.             out[0][n][1] = (UINTFLOAT)in[6][n][0] + in[7][n][0];

  177.             out[1][n][1] = (UINTFLOAT)in[6][n][1] + in[7][n][1];

  178.             out[0][n][2] = (UINTFLOAT)in[8][n][0] + in[9][n][0];

  179.             out[1][n][2] = (UINTFLOAT)in[8][n][1] + in[9][n][1];

  180.         }

  181.         dsp->hybrid_synthesis_deint(out, in + 7, 3, len);

  182.     }

  183. }

  184. 

  185. /// All-pass filter decay slope

  186. #define DECAY_SLOPE      Q30(0.05f)

  187. /// Number of frequency bands that can be addressed by the parameter index, b(k)

  188. static const int   NR_PAR_BANDS[]      = { 20, 34 };

  189. static const int   NR_IPDOPD_BANDS[]   = { 11, 17 };

  190. /// Number of frequency bands that can be addressed by the sub subband index, k

  191. static const int   NR_BANDS[]          = { 71, 91 };

  192. /// Start frequency band for the all-pass filter decay slope

  193. static const int   DECAY_CUTOFF[]      = { 10, 32 };

  194. /// Number of all-pass filer bands

  195. static const int   NR_ALLPASS_BANDS[]  = { 30, 50 };

  196. /// First stereo band using the short one sample delay

  197. static const int   SHORT_DELAY_BAND[]  = { 42, 62 };

  198. 

  199. /** Table 8.46 */

  200. static void map_idx_10_to_20(int8_t *par_mapped, const int8_t *par, int full)

  201. {

  202.     int b;

  203.     if (full)

  204.         b = 9;

  205.     else {

  206.         b = 4;

  207.         par_mapped[10] = 0;

  208.     }

  209.     for (; b >= 0; b--) {

  210.         par_mapped[2*b+1] = par_mapped[2*b] = par[b];

  211.     }

  212. }

  213. 

  214. static void map_idx_34_to_20(int8_t *par_mapped, const int8_t *par, int full)

  215. {

  216.     par_mapped[ 0] = (2*par[ 0] +   par[ 1]) / 3;

  217.     par_mapped[ 1] = (  par[ 1] + 2*par[ 2]) / 3;

  218.     par_mapped[ 2] = (2*par[ 3] +   par[ 4]) / 3;

  219.     par_mapped[ 3] = (  par[ 4] + 2*par[ 5]) / 3;

  220.     par_mapped[ 4] = (  par[ 6] +   par[ 7]) / 2;

  221.     par_mapped[ 5] = (  par[ 8] +   par[ 9]) / 2;

  222.     par_mapped[ 6] =    par[10];

  223.     par_mapped[ 7] =    par[11];

  224.     par_mapped[ 8] = (  par[12] +   par[13]) / 2;

  225.     par_mapped[ 9] = (  par[14] +   par[15]) / 2;

  226.     par_mapped[10] =    par[16];

  227.     if (full) {

  228.         par_mapped[11] =    par[17];

  229.         par_mapped[12] =    par[18];

  230.         par_mapped[13] =    par[19];

  231.         par_mapped[14] = (  par[20] +   par[21]) / 2;

  232.         par_mapped[15] = (  par[22] +   par[23]) / 2;

  233.         par_mapped[16] = (  par[24] +   par[25]) / 2;

  234.         par_mapped[17] = (  par[26] +   par[27]) / 2;

  235.         par_mapped[18] = (  par[28] +   par[29] +   par[30] +   par[31]) / 4;

  236.         par_mapped[19] = (  par[32] +   par[33]) / 2;

  237.     }

  238. }

  239. 

  240. static void map_val_34_to_20(INTFLOAT par[PS_MAX_NR_IIDICC])

  241. {

  242. #if USE_FIXED

  243.     par[ 0] = (int)(((int64_t)(par[ 0] + (unsigned)(par[ 1]>>1)) * 1431655765 + \

  244.                       0x40000000) >> 31);

  245.     par[ 1] = (int)(((int64_t)((par[ 1]>>1) + (unsigned)par[ 2]) * 1431655765 + \

  246.                       0x40000000) >> 31);

  247.     par[ 2] = (int)(((int64_t)(par[ 3] + (unsigned)(par[ 4]>>1)) * 1431655765 + \

  248.                       0x40000000) >> 31);

  249.     par[ 3] = (int)(((int64_t)((par[ 4]>>1) + (unsigned)par[ 5]) * 1431655765 + \

  250.                       0x40000000) >> 31);

  251. #else

  252.     par[ 0] = (2*par[ 0] +   par[ 1]) * 0.33333333f;

  253.     par[ 1] = (  par[ 1] + 2*par[ 2]) * 0.33333333f;

  254.     par[ 2] = (2*par[ 3] +   par[ 4]) * 0.33333333f;

  255.     par[ 3] = (  par[ 4] + 2*par[ 5]) * 0.33333333f;

  256. #endif /* USE_FIXED */

  257.     par[ 4] = AAC_HALF_SUM(par[ 6], par[ 7]);

  258.     par[ 5] = AAC_HALF_SUM(par[ 8], par[ 9]);

  259.     par[ 6] =    par[10];

  260.     par[ 7] =    par[11];

  261.     par[ 8] = AAC_HALF_SUM(par[12], par[13]);

  262.     par[ 9] = AAC_HALF_SUM(par[14], par[15]);

  263.     par[10] =    par[16];

  264.     par[11] =    par[17];

  265.     par[12] =    par[18];

  266.     par[13] =    par[19];

  267.     par[14] = AAC_HALF_SUM(par[20], par[21]);

  268.     par[15] = AAC_HALF_SUM(par[22], par[23]);

  269.     par[16] = AAC_HALF_SUM(par[24], par[25]);

  270.     par[17] = AAC_HALF_SUM(par[26], par[27]);

  271. #if USE_FIXED

  272.     par[18] = (((par[28]+2)>>2) + ((par[29]+2)>>2) + ((par[30]+2)>>2) + ((par[31]+2)>>2));

  273. #else

  274.     par[18] = (  par[28] +   par[29] +   par[30] +   par[31]) * 0.25f;

  275. #endif /* USE_FIXED */

  276.     par[19] = AAC_HALF_SUM(par[32], par[33]);

  277. }

  278. 

  279. static void map_idx_10_to_34(int8_t *par_mapped, const int8_t *par, int full)

  280. {

  281.     if (full) {

  282.         par_mapped[33] = par[9];

  283.         par_mapped[32] = par[9];

  284.         par_mapped[31] = par[9];

  285.         par_mapped[30] = par[9];

  286.         par_mapped[29] = par[9];

  287.         par_mapped[28] = par[9];

  288.         par_mapped[27] = par[8];

  289.         par_mapped[26] = par[8];

  290.         par_mapped[25] = par[8];

  291.         par_mapped[24] = par[8];

  292.         par_mapped[23] = par[7];

  293.         par_mapped[22] = par[7];

  294.         par_mapped[21] = par[7];

  295.         par_mapped[20] = par[7];

  296.         par_mapped[19] = par[6];

  297.         par_mapped[18] = par[6];

  298.         par_mapped[17] = par[5];

  299.         par_mapped[16] = par[5];

  300.     } else {

  301.         par_mapped[16] =      0;

  302.     }

  303.     par_mapped[15] = par[4];

  304.     par_mapped[14] = par[4];

  305.     par_mapped[13] = par[4];

  306.     par_mapped[12] = par[4];

  307.     par_mapped[11] = par[3];

  308.     par_mapped[10] = par[3];

  309.     par_mapped[ 9] = par[2];

  310.     par_mapped[ 8] = par[2];

  311.     par_mapped[ 7] = par[2];

  312.     par_mapped[ 6] = par[2];

  313.     par_mapped[ 5] = par[1];

  314.     par_mapped[ 4] = par[1];

  315.     par_mapped[ 3] = par[1];

  316.     par_mapped[ 2] = par[0];

  317.     par_mapped[ 1] = par[0];

  318.     par_mapped[ 0] = par[0];

  319. }

  320. 

  321. static void map_idx_20_to_34(int8_t *par_mapped, const int8_t *par, int full)

  322. {

  323.     if (full) {

  324.         par_mapped[33] =  par[19];

  325.         par_mapped[32] =  par[19];

  326.         par_mapped[31] =  par[18];

  327.         par_mapped[30] =  par[18];

  328.         par_mapped[29] =  par[18];

  329.         par_mapped[28] =  par[18];

  330.         par_mapped[27] =  par[17];

  331.         par_mapped[26] =  par[17];

  332.         par_mapped[25] =  par[16];

  333.         par_mapped[24] =  par[16];

  334.         par_mapped[23] =  par[15];

  335.         par_mapped[22] =  par[15];

  336.         par_mapped[21] =  par[14];

  337.         par_mapped[20] =  par[14];

  338.         par_mapped[19] =  par[13];

  339.         par_mapped[18] =  par[12];

  340.         par_mapped[17] =  par[11];

  341.     }

  342.     par_mapped[16] =  par[10];

  343.     par_mapped[15] =  par[ 9];

  344.     par_mapped[14] =  par[ 9];

  345.     par_mapped[13] =  par[ 8];

  346.     par_mapped[12] =  par[ 8];

  347.     par_mapped[11] =  par[ 7];

  348.     par_mapped[10] =  par[ 6];

  349.     par_mapped[ 9] =  par[ 5];

  350.     par_mapped[ 8] =  par[ 5];

  351.     par_mapped[ 7] =  par[ 4];

  352.     par_mapped[ 6] =  par[ 4];

  353.     par_mapped[ 5] =  par[ 3];

  354.     par_mapped[ 4] = (par[ 2] + par[ 3]) / 2;

  355.     par_mapped[ 3] =  par[ 2];

  356.     par_mapped[ 2] =  par[ 1];

  357.     par_mapped[ 1] = (par[ 0] + par[ 1]) / 2;

  358.     par_mapped[ 0] =  par[ 0];

  359. }

  360. 

  361. static void map_val_20_to_34(INTFLOAT par[PS_MAX_NR_IIDICC])

  362. {

  363.     par[33] =  par[19];

  364.     par[32] =  par[19];

  365.     par[31] =  par[18];

  366.     par[30] =  par[18];

  367.     par[29] =  par[18];

  368.     par[28] =  par[18];

  369.     par[27] =  par[17];

  370.     par[26] =  par[17];

  371.     par[25] =  par[16];

  372.     par[24] =  par[16];

  373.     par[23] =  par[15];

  374.     par[22] =  par[15];

  375.     par[21] =  par[14];

  376.     par[20] =  par[14];

  377.     par[19] =  par[13];

  378.     par[18] =  par[12];

  379.     par[17] =  par[11];

  380.     par[16] =  par[10];

  381.     par[15] =  par[ 9];

  382.     par[14] =  par[ 9];

  383.     par[13] =  par[ 8];

  384.     par[12] =  par[ 8];

  385.     par[11] =  par[ 7];

  386.     par[10] =  par[ 6];

  387.     par[ 9] =  par[ 5];

  388.     par[ 8] =  par[ 5];

  389.     par[ 7] =  par[ 4];

  390.     par[ 6] =  par[ 4];

  391.     par[ 5] =  par[ 3];

  392.     par[ 4] = AAC_HALF_SUM(par[ 2], par[ 3]);

  393.     par[ 3] =  par[ 2];

  394.     par[ 2] =  par[ 1];

  395.     par[ 1] = AAC_HALF_SUM(par[ 0], par[ 1]);

  396. }

  397. 

  398. static void decorrelation(PSContext *ps, INTFLOAT (*out)[32][2], const INTFLOAT (*s)[32][2], int is34)

  399. {

  400.     LOCAL_ALIGNED_16(INTFLOAT, power, [34], [PS_QMF_TIME_SLOTS]);

  401.     LOCAL_ALIGNED_16(INTFLOAT, transient_gain, [34], [PS_QMF_TIME_SLOTS]);

  402.     INTFLOAT *peak_decay_nrg = ps->peak_decay_nrg;

  403.     INTFLOAT *power_smooth = ps->power_smooth;

  404.     INTFLOAT *peak_decay_diff_smooth = ps->peak_decay_diff_smooth;

  405.     INTFLOAT (*delay)[PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2] = ps->delay;

  406.     INTFLOAT (*ap_delay)[PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2] = ps->ap_delay;

  407. #if !USE_FIXED

  408.     const float transient_impact  = 1.5f;

  409.     const float a_smooth          = 0.25f; ///< Smoothing coefficient

  410. #endif /* USE_FIXED */

  411.     const int8_t *const k_to_i = is34 ? ff_k_to_i_34 : ff_k_to_i_20;

  412.     int i, k, m, n;

  413.     int n0 = 0, nL = 32;

  414.     const INTFLOAT peak_decay_factor = Q31(0.76592833836465f);

  415. 

  416.     memset(power, 0, 34 * sizeof(*power));

  417. 

  418.     if (is34 != ps->common.is34bands_old) {

  419.         memset(ps->peak_decay_nrg,         0, sizeof(ps->peak_decay_nrg));

  420.         memset(ps->power_smooth,           0, sizeof(ps->power_smooth));

  421.         memset(ps->peak_decay_diff_smooth, 0, sizeof(ps->peak_decay_diff_smooth));

  422.         memset(ps->delay,                  0, sizeof(ps->delay));

  423.         memset(ps->ap_delay,               0, sizeof(ps->ap_delay));

  424.     }

  425. 

  426.     for (k = 0; k < NR_BANDS[is34]; k++) {

  427.         int i = k_to_i[k];

  428.         ps->dsp.add_squares(power[i], s[k], nL - n0);

  429.     }

  430. 

  431.     //Transient detection

  432. #if USE_FIXED

  433.     for (i = 0; i < NR_PAR_BANDS[is34]; i++) {

  434.         for (n = n0; n < nL; n++) {

  435.             int decayed_peak;

  436.             decayed_peak = (int)(((int64_t)peak_decay_factor * \

  437.                                            peak_decay_nrg[i] + 0x40000000) >> 31);

  438.             peak_decay_nrg[i] = FFMAX(decayed_peak, power[i][n]);

  439.             power_smooth[i] += (power[i][n] + 2LL - power_smooth[i]) >> 2;

  440.             peak_decay_diff_smooth[i] += (peak_decay_nrg[i] + 2LL - power[i][n] - \

  441.                                           peak_decay_diff_smooth[i]) >> 2;

  442. 

  443.             if (peak_decay_diff_smooth[i]) {

  444.                 transient_gain[i][n] = FFMIN(power_smooth[i]*43691LL / peak_decay_diff_smooth[i], 1<<16);

  445.             } else

  446.                 transient_gain[i][n] = 1 << 16;

  447.         }

  448.     }

  449. #else

  450.     for (i = 0; i < NR_PAR_BANDS[is34]; i++) {

  451.         for (n = n0; n < nL; n++) {

  452.             float decayed_peak = peak_decay_factor * peak_decay_nrg[i];

  453.             float denom;

  454.             peak_decay_nrg[i] = FFMAX(decayed_peak, power[i][n]);

  455.             power_smooth[i] += a_smooth * (power[i][n] - power_smooth[i]);

  456.             peak_decay_diff_smooth[i] += a_smooth * (peak_decay_nrg[i] - power[i][n] - peak_decay_diff_smooth[i]);

  457.             denom = transient_impact * peak_decay_diff_smooth[i];

  458.             transient_gain[i][n]   = (denom > power_smooth[i]) ?

  459.                                          power_smooth[i] / denom : 1.0f;

  460.         }

  461.     }

  462. 

  463. #endif /* USE_FIXED */

  464.     //Decorrelation and transient reduction

  465.     //                         PS_AP_LINKS - 1

  466.     //                               -----

  467.     //                                | |  Q_fract_allpass[k][m]*z^-link_delay[m] - a[m]*g_decay_slope[k]

  468.     //H[k][z] = z^-2 * phi_fract[k] * | | ----------------------------------------------------------------

  469.     //                                | | 1 - a[m]*g_decay_slope[k]*Q_fract_allpass[k][m]*z^-link_delay[m]

  470.     //                               m = 0

  471.     //d[k][z] (out) = transient_gain_mapped[k][z] * H[k][z] * s[k][z]

  472.     for (k = 0; k < NR_ALLPASS_BANDS[is34]; k++) {

  473.         int b = k_to_i[k];

  474. #if USE_FIXED

  475.         int g_decay_slope;

  476. 

  477.         if (k - DECAY_CUTOFF[is34] <= 0) {

  478.           g_decay_slope = 1 << 30;

  479.         }

  480.         else if (k - DECAY_CUTOFF[is34] >= 20) {

  481.           g_decay_slope = 0;

  482.         }

  483.         else {

  484.           g_decay_slope = (1 << 30) - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]);

  485.         }

  486. #else

  487.         float g_decay_slope = 1.f - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]);

  488.         g_decay_slope = av_clipf(g_decay_slope, 0.f, 1.f);

  489. #endif /* USE_FIXED */

  490.         memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));

  491.         memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));

  492.         for (m = 0; m < PS_AP_LINKS; m++) {

  493.             memcpy(ap_delay[k][m],   ap_delay[k][m]+numQMFSlots,           5*sizeof(ap_delay[k][m][0]));

  494.         }

  495.         ps->dsp.decorrelate(out[k], delay[k] + PS_MAX_DELAY - 2, ap_delay[k],

  496.                             phi_fract[is34][k],

  497.                             (const INTFLOAT (*)[2]) Q_fract_allpass[is34][k],

  498.                             transient_gain[b], g_decay_slope, nL - n0);

  499.     }

  500.     for (; k < SHORT_DELAY_BAND[is34]; k++) {

  501.         int i = k_to_i[k];

  502.         memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));

  503.         memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));

  504.         //H = delay 14

  505.         ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 14,

  506.                                 transient_gain[i], nL - n0);

  507.     }

  508.     for (; k < NR_BANDS[is34]; k++) {

  509.         int i = k_to_i[k];

  510.         memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));

  511.         memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));

  512.         //H = delay 1

  513.         ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 1,

  514.                                 transient_gain[i], nL - n0);

  515.     }

  516. }

  517. 

  518. static void remap34(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC],

  519.                     int8_t           (*par)[PS_MAX_NR_IIDICC],

  520.                     int num_par, int num_env, int full)

  521. {

  522.     int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped;

  523.     int e;

  524.     if (num_par == 20 || num_par == 11) {

  525.         for (e = 0; e < num_env; e++) {

  526.             map_idx_20_to_34(par_mapped[e], par[e], full);

  527.         }

  528.     } else if (num_par == 10 || num_par == 5) {

  529.         for (e = 0; e < num_env; e++) {

  530.             map_idx_10_to_34(par_mapped[e], par[e], full);

  531.         }

  532.     } else {

  533.         *p_par_mapped = par;

  534.     }

  535. }

  536. 

  537. static void remap20(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC],

  538.                     int8_t           (*par)[PS_MAX_NR_IIDICC],

  539.                     int num_par, int num_env, int full)

  540. {

  541.     int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped;

  542.     int e;

  543.     if (num_par == 34 || num_par == 17) {

  544.         for (e = 0; e < num_env; e++) {

  545.             map_idx_34_to_20(par_mapped[e], par[e], full);

  546.         }

  547.     } else if (num_par == 10 || num_par == 5) {

  548.         for (e = 0; e < num_env; e++) {

  549.             map_idx_10_to_20(par_mapped[e], par[e], full);

  550.         }

  551.     } else {

  552.         *p_par_mapped = par;

  553.     }

  554. }

  555. 

  556. static void stereo_processing(PSContext *ps, INTFLOAT (*l)[32][2], INTFLOAT (*r)[32][2], int is34)

  557. {

  558.     int e, b, k;

  559. 

  560.     PSCommonContext *const ps2 = &ps->common;

  561.     INTFLOAT (*H11)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H11;

  562.     INTFLOAT (*H12)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H12;

  563.     INTFLOAT (*H21)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H21;

  564.     INTFLOAT (*H22)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H22;

  565.     int8_t *opd_hist = ps->opd_hist;

  566.     int8_t *ipd_hist = ps->ipd_hist;

  567.     int8_t iid_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];

  568.     int8_t icc_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];

  569.     int8_t ipd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];

  570.     int8_t opd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];

  571.     int8_t (*iid_mapped)[PS_MAX_NR_IIDICC] = iid_mapped_buf;

  572.     int8_t (*icc_mapped)[PS_MAX_NR_IIDICC] = icc_mapped_buf;

  573.     int8_t (*ipd_mapped)[PS_MAX_NR_IIDICC] = ipd_mapped_buf;

  574.     int8_t (*opd_mapped)[PS_MAX_NR_IIDICC] = opd_mapped_buf;

  575.     const int8_t *const k_to_i = is34 ? ff_k_to_i_34 : ff_k_to_i_20;

  576.     TABLE_CONST INTFLOAT (*H_LUT)[8][4] = (PS_BASELINE || ps2->icc_mode < 3) ? HA : HB;

  577. 

  578.     //Remapping

  579.     if (ps2->num_env_old) {

  580.         memcpy(H11[0][0], H11[0][ps2->num_env_old], sizeof(H11[0][0]));

  581.         memcpy(H11[1][0], H11[1][ps2->num_env_old], sizeof(H11[1][0]));

  582.         memcpy(H12[0][0], H12[0][ps2->num_env_old], sizeof(H12[0][0]));

  583.         memcpy(H12[1][0], H12[1][ps2->num_env_old], sizeof(H12[1][0]));

  584.         memcpy(H21[0][0], H21[0][ps2->num_env_old], sizeof(H21[0][0]));

  585.         memcpy(H21[1][0], H21[1][ps2->num_env_old], sizeof(H21[1][0]));

  586.         memcpy(H22[0][0], H22[0][ps2->num_env_old], sizeof(H22[0][0]));

  587.         memcpy(H22[1][0], H22[1][ps2->num_env_old], sizeof(H22[1][0]));

  588.     }

  589. 

  590.     if (is34) {

  591.         remap34(&iid_mapped, ps2->iid_par, ps2->nr_iid_par, ps2->num_env, 1);

  592.         remap34(&icc_mapped, ps2->icc_par, ps2->nr_icc_par, ps2->num_env, 1);

  593.         if (ps2->enable_ipdopd) {

  594.             remap34(&ipd_mapped, ps2->ipd_par, ps2->nr_ipdopd_par, ps2->num_env, 0);

  595.             remap34(&opd_mapped, ps2->opd_par, ps2->nr_ipdopd_par, ps2->num_env, 0);

  596.         }

  597.         if (!ps2->is34bands_old) {

  598.             map_val_20_to_34(H11[0][0]);

  599.             map_val_20_to_34(H11[1][0]);

  600.             map_val_20_to_34(H12[0][0]);

  601.             map_val_20_to_34(H12[1][0]);

  602.             map_val_20_to_34(H21[0][0]);

  603.             map_val_20_to_34(H21[1][0]);

  604.             map_val_20_to_34(H22[0][0]);

  605.             map_val_20_to_34(H22[1][0]);

  606.             ipdopd_reset(ipd_hist, opd_hist);

  607.         }

  608.     } else {

  609.         remap20(&iid_mapped, ps2->iid_par, ps2->nr_iid_par, ps2->num_env, 1);

  610.         remap20(&icc_mapped, ps2->icc_par, ps2->nr_icc_par, ps2->num_env, 1);

  611.         if (ps2->enable_ipdopd) {

  612.             remap20(&ipd_mapped, ps2->ipd_par, ps2->nr_ipdopd_par, ps2->num_env, 0);

  613.             remap20(&opd_mapped, ps2->opd_par, ps2->nr_ipdopd_par, ps2->num_env, 0);

  614.         }

  615.         if (ps2->is34bands_old) {

  616.             map_val_34_to_20(H11[0][0]);

  617.             map_val_34_to_20(H11[1][0]);

  618.             map_val_34_to_20(H12[0][0]);

  619.             map_val_34_to_20(H12[1][0]);

  620.             map_val_34_to_20(H21[0][0]);

  621.             map_val_34_to_20(H21[1][0]);

  622.             map_val_34_to_20(H22[0][0]);

  623.             map_val_34_to_20(H22[1][0]);

  624.             ipdopd_reset(ipd_hist, opd_hist);

  625.         }

  626.     }

  627. 

  628.     //Mixing

  629.     for (e = 0; e < ps2->num_env; e++) {

  630.         for (b = 0; b < NR_PAR_BANDS[is34]; b++) {

  631.             INTFLOAT h11, h12, h21, h22;

  632.             h11 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps2->iid_quant][icc_mapped[e][b]][0];

  633.             h12 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps2->iid_quant][icc_mapped[e][b]][1];

  634.             h21 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps2->iid_quant][icc_mapped[e][b]][2];

  635.             h22 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps2->iid_quant][icc_mapped[e][b]][3];

  636. 

  637.             if (!PS_BASELINE && ps2->enable_ipdopd && b < NR_IPDOPD_BANDS[is34]) {

  638.                 //The spec say says to only run this smoother when enable_ipdopd

  639.                 //is set but the reference decoder appears to run it constantly

  640.                 INTFLOAT h11i, h12i, h21i, h22i;

  641.                 INTFLOAT ipd_adj_re, ipd_adj_im;

  642.                 int opd_idx = opd_hist[b] * 8 + opd_mapped[e][b];

  643.                 int ipd_idx = ipd_hist[b] * 8 + ipd_mapped[e][b];

  644.                 INTFLOAT opd_re = pd_re_smooth[opd_idx];

  645.                 INTFLOAT opd_im = pd_im_smooth[opd_idx];

  646.                 INTFLOAT ipd_re = pd_re_smooth[ipd_idx];

  647.                 INTFLOAT ipd_im = pd_im_smooth[ipd_idx];

  648.                 opd_hist[b] = opd_idx & 0x3F;

  649.                 ipd_hist[b] = ipd_idx & 0x3F;

  650. 

  651.                 ipd_adj_re = AAC_MADD30(opd_re, ipd_re, opd_im, ipd_im);

  652.                 ipd_adj_im = AAC_MSUB30(opd_im, ipd_re, opd_re, ipd_im);

  653.                 h11i = AAC_MUL30(h11,  opd_im);

  654.                 h11  = AAC_MUL30(h11,  opd_re);

  655.                 h12i = AAC_MUL30(h12,  ipd_adj_im);

  656.                 h12  = AAC_MUL30(h12,  ipd_adj_re);

  657.                 h21i = AAC_MUL30(h21,  opd_im);

  658.                 h21  = AAC_MUL30(h21,  opd_re);

  659.                 h22i = AAC_MUL30(h22,  ipd_adj_im);

  660.                 h22  = AAC_MUL30(h22,  ipd_adj_re);

  661.                 H11[1][e+1][b] = h11i;

  662.                 H12[1][e+1][b] = h12i;

  663.                 H21[1][e+1][b] = h21i;

  664.                 H22[1][e+1][b] = h22i;

  665.             }

  666.             H11[0][e+1][b] = h11;

  667.             H12[0][e+1][b] = h12;

  668.             H21[0][e+1][b] = h21;

  669.             H22[0][e+1][b] = h22;

  670.         }

  671.         for (k = 0; k < NR_BANDS[is34]; k++) {

  672.             LOCAL_ALIGNED_16(INTFLOAT, h, [2], [4]);

  673.             LOCAL_ALIGNED_16(INTFLOAT, h_step, [2], [4]);

  674.             int start = ps2->border_position[e];

  675.             int stop  = ps2->border_position[e+1];

  676.             INTFLOAT width = Q30(1.f) / ((stop - start) ? (stop - start) : 1);

  677. #if USE_FIXED

  678.             width = FFMIN(2U*width, INT_MAX);

  679. #endif

  680.             b = k_to_i[k];

  681.             h[0][0] = H11[0][e][b];

  682.             h[0][1] = H12[0][e][b];

  683.             h[0][2] = H21[0][e][b];

  684.             h[0][3] = H22[0][e][b];

  685.             if (!PS_BASELINE && ps2->enable_ipdopd) {

  686.             //Is this necessary? ps_04_new seems unchanged

  687.             if ((is34 && k <= 13 && k >= 9) || (!is34 && k <= 1)) {

  688.                 h[1][0] = -H11[1][e][b];

  689.                 h[1][1] = -H12[1][e][b];

  690.                 h[1][2] = -H21[1][e][b];

  691.                 h[1][3] = -H22[1][e][b];

  692.             } else {

  693.                 h[1][0] = H11[1][e][b];

  694.                 h[1][1] = H12[1][e][b];

  695.                 h[1][2] = H21[1][e][b];

  696.                 h[1][3] = H22[1][e][b];

  697.             }

  698.             }

  699.             //Interpolation

  700.             h_step[0][0] = AAC_MSUB31_V3(H11[0][e+1][b], h[0][0], width);

  701.             h_step[0][1] = AAC_MSUB31_V3(H12[0][e+1][b], h[0][1], width);

  702.             h_step[0][2] = AAC_MSUB31_V3(H21[0][e+1][b], h[0][2], width);

  703.             h_step[0][3] = AAC_MSUB31_V3(H22[0][e+1][b], h[0][3], width);

  704.             if (!PS_BASELINE && ps2->enable_ipdopd) {

  705.                 h_step[1][0] = AAC_MSUB31_V3(H11[1][e+1][b], h[1][0], width);

  706.                 h_step[1][1] = AAC_MSUB31_V3(H12[1][e+1][b], h[1][1], width);

  707.                 h_step[1][2] = AAC_MSUB31_V3(H21[1][e+1][b], h[1][2], width);

  708.                 h_step[1][3] = AAC_MSUB31_V3(H22[1][e+1][b], h[1][3], width);

  709.             }

  710.             if (stop - start)

  711.                 ps->dsp.stereo_interpolate[!PS_BASELINE && ps2->enable_ipdopd](

  712.                     l[k] + 1 + start, r[k] + 1 + start,

  713.                     h, h_step, stop - start);

  714.         }

  715.     }

  716. }

  717. 

  718. int AAC_RENAME(ff_ps_apply)(AVCodecContext *avctx, PSContext *ps, INTFLOAT L[2][38][64], INTFLOAT R[2][38][64], int top)

  719. {

  720.     INTFLOAT (*Lbuf)[32][2] = ps->Lbuf;

  721.     INTFLOAT (*Rbuf)[32][2] = ps->Rbuf;

  722.     const int len = 32;

  723.     int is34 = ps->common.is34bands;

  724. 

  725.     top += NR_BANDS[is34] - 64;

  726.     memset(ps->delay+top, 0, (NR_BANDS[is34] - top)*sizeof(ps->delay[0]));

  727.     if (top < NR_ALLPASS_BANDS[is34])

  728.         memset(ps->ap_delay + top, 0, (NR_ALLPASS_BANDS[is34] - top)*sizeof(ps->ap_delay[0]));

  729. 

  730.     hybrid_analysis(&ps->dsp, Lbuf, ps->in_buf, L, is34, len);

  731.     decorrelation(ps, Rbuf, (const INTFLOAT (*)[32][2]) Lbuf, is34);

  732.     stereo_processing(ps, Lbuf, Rbuf, is34);

  733.     hybrid_synthesis(&ps->dsp, L, Lbuf, is34, len);

  734.     hybrid_synthesis(&ps->dsp, R, Rbuf, is34, len);

  735. 

  736.     return 0;

  737. }

  738. 

  739. av_cold void AAC_RENAME(ff_ps_init)(void) {

  740.     ps_tableinit();

  741.     ff_ps_init_common();

  742. }

  743. 

  744. av_cold void AAC_RENAME(ff_ps_ctx_init)(PSContext *ps)

  745. {

  746.     AAC_RENAME(ff_psdsp_init)(&ps->dsp);

  747. }